Unified Dual-Intent Translation for Joint Modeling of Search and Recommendation
统一双意图翻译:搜索与推荐联合建模
Abstract. 摘要。
Recommendation systems, which assist users in discovering their preferred items among numerous options, have served billions of users across various online platforms. Intuitively, users’ interactions with items are highly driven by their unchanging inherent intents (e.g., always preferring high-quality items) and changing demand intents (e.g., wanting a T-shirt in summer but a down jacket in winter). However, both types of intents are implicitly expressed in recommendation scenario, posing challenges in leveraging them for accurate intent-aware recommendations. Fortunately, in search scenario, often found alongside recommendation on the same online platform, users express their demand intents explicitly through their query words. Intuitively, in both scenarios, a user shares the same inherent intent and his/her interactions may be influenced by the same demand intent. It is therefore feasible to utilize the interaction data from both scenarios to reinforce the dual intents for joint intent-aware modeling. But the joint modeling should deal with two problems: (1) accurately modeling users’ implicit demand intents in recommendation; (2) modeling the relation between the dual intents and the interactive items. To address these problems, we propose a novel model named Unified Dual-Intents Translation for joint modeling of Search and Recommendation (UDITSR). To accurately simulate users’ demand intents in recommendation, we utilize real queries from search data as supervision information to guide its generation. To explicitly model the relation among the triplet ¡inherent intent, demand intent, interactive item¿, we propose a dual-intent translation propagation mechanism to learn the triplet in the same semantic space via embedding translations. Extensive experiments demonstrate that UDITSR outperforms SOTA baselines both in search and recommendation tasks. Moreover, our model has been deployed online on Meituan Waimai platform, leading to an average improvement in GMV (Gross Merchandise Value) of 1.46% and CTR(Click-Through Rate) of 0.77% over one month.
推荐系统帮助用户在众多选项中发掘其偏好的物品,已服务于各大在线平台的数十亿用户。直观来看,用户与物品的互动深受其不变的内在意图(如始终偏好高质量物品)和变化的需求意图(如夏季想要 T 恤而冬季则需要羽绒服)的影响。然而,这两种意图在推荐场景中均隐性表达,给利用它们进行精准的意图感知推荐带来了挑战。幸运的是,在搜索场景中,通常与推荐并存于同一在线平台,用户通过查询词明确表达其需求意图。直观上,在这两种场景中,用户共享相同的内在意图,其互动可能受相同的需求意图影响。因此,利用两种场景的互动数据来强化双重意图,进行联合意图感知建模是可行的。 然而,联合建模需解决两大问题:(1)精准建模用户在推荐中的隐性需求意图;(2)建模双重意图与交互项之间的关系。为应对这些问题,我们提出了一种名为“搜索与推荐联合建模的统一双重意图转换模型”(UDITSR)的新模型。为精准模拟用户在推荐中的需求意图,我们利用搜索数据中的真实查询作为监督信息来引导其生成。为明确建模“固有意图、需求意图、交互项”三元组之间的关系,我们提出了一种双重意图转换传播机制,通过嵌入转换在同一语义空间中学习该三元组。大量实验证明,UDITSR 在搜索与推荐任务中均优于现有最先进基线。此外,我们的模型已在美团外卖平台上线部署,一个月内平均提升了 1.46%的 GMV(商品交易总额)和 0.77%的 CTR(点击通过率)。
联合学习,搜索与推荐,双重意图建模,意图转换
† 版权:保留所有权利††conference: Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining; August 25–29, 2024; Barcelona, Spain
† 会议:第 30 届 ACM SIGKDD 知识发现与数据挖掘会议论文集;2024 年 8 月 25 日至 29 日;西班牙巴塞罗那††booktitle: Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD ’24), August 25–29, 2024, Barcelona, Spain
† 书名:第 30 届 ACM SIGKDD 知识发现与数据挖掘会议(KDD '24)论文集,2024 年 8 月 25 日至 29 日,西班牙巴塞罗那††doi: 10.1145/3637528.3671519††isbn: 979-8-4007-0490-1/24/08††ccs: Information systems Recommender systems
† 信息系统 推荐系统
1. Introduction 1. 引言
Aiming to help users discover items of interest from a vast array of options,
recommendation systems have become an essential component of various online platforms, such as e-commerce (Zhou et al., 2018, 2019; Pi et al., 2020) and digital news services (Li et al., 2010; Covington et al., 2016; Tai et al., 2021). Existing recommendation models (Hu et al., 2008; Zhou et al., 2018; He et al., 2017; Zhou et al., 2019) typically exploit users’ implicit feedback, such as click history, to predict their interests. For instance, traditional Collaborative Filtering (CF) (Hu et al., 2008) assumes that users will interact with items similar to those with which they’ve previously interacted. Furthermore, various models (Zhou et al., 2018, 2019) have been developed to capture the sequential dynamics of users’ implicit feedback to model their evolving interests.
旨在帮助用户从海量选项中发现感兴趣的物品,推荐系统已成为各类在线平台不可或缺的组成部分,如电子商务(Zhou 等,2018,2019;Pi 等,2020)和数字新闻服务(Li 等,2010;Covington 等,2016;Tai 等,2021)。现有的推荐模型(Hu 等,2008;Zhou 等,2018;He 等,2017;Zhou 等,2019)通常利用用户的隐式反馈,如点击历史,来预测他们的兴趣。例如,传统的协同过滤(CF)(Hu 等,2008)假设用户会与那些与他们之前互动过的物品相似的物品进行互动。此外,多种模型(Zhou 等,2018,2019)已被开发出来,以捕捉用户隐式反馈的序列动态,从而建模他们不断变化的兴趣。
In practice, user feedback patterns in recommendation systems are highly driven by their complex intents, which can be broadly categorized into unchanging inherent intents and changing demand intents. For example, Amy and Tom may have the same noodle demand but choose different restaurants due to Amy’s inherent intent for spicy flavors and Tom’s for sweet. Besides, a single user’s interactions can vary due to their changing demands. Yet, these intents are often implicitly expressed in the recommendation, presenting a challenge for accurate intent-aware recommendations.
Existing intent-aware recommendation models (Chen et al., 2019; Zhu et al., 2020; Liu et al., 2020b) typically rely on users’ implicit feedback to learn their intents. However, these models encounter a significant problem: different users may have different inherent or demand intents despite similar historical feedback. As shown in Figure 1(a), Amy’s interaction with Pizza Hut might indicate a demand intent for pasta, while Tom may demand pizza instead.
Ideally, recommendation systems should suggest pasta-related options to Amy and pizza-related ones to Tom. However, without any explicit intent information, existing models struggle to distinguish between these intents, resulting in inaccurate recommendations.
在实际应用中,推荐系统中的用户反馈模式深受其复杂意图的驱动,这些意图大致可分为不变的固有意图和变化的需求意图。例如,Amy 和 Tom 可能有相同的吃面需求,但由于 Amy 偏爱辣味而 Tom 偏好甜味,他们选择了不同的餐厅。此外,单个用户的交互行为也会因需求变化而有所不同。然而,这些意图往往在推荐中隐含表达,给精准的意图感知推荐带来了挑战。现有的意图感知推荐模型(如 Chen 等,2019;Zhu 等,2020;Liu 等,2020b)通常依赖用户的隐式反馈来学习其意图。然而,这些模型面临一个重大问题:尽管历史反馈相似,不同用户可能拥有不同的固有或需求意图。如图 1(a)所示,Amy 与必胜客的互动可能表明她对意面的需求意图,而 Tom 则可能需求披萨。理想情况下,推荐系统应向 Amy 推荐意面相关选项,向 Tom 推荐披萨相关选项。 然而,在没有明确意图信息的情况下,现有模型难以区分这些意图,导致推荐结果不准确。
Fortunately, in search services, which often accompany recommendation services on the same online platform, users explicitly express their demand intents through query words, as shown in Figure 1(b). Such explicit search demand information can serve as additional explicit information to assist in learning implicit demand intents for recommendation. Indeed, both search and recommendation tasks aim to comprehend users’ intents to aid them in obtaining desired items (Belkin and Croft, 1992).
In addition, in search scenario, users’ interactions are influenced not only by their explicit demand intents but also by their personalized inherent intents. Yet, search models typically focus on the match between search results and users’ demand intents, often overlooking the impact of their personalized inherent intents, which are indeed significant (Sondhi et al., 2018). Intuitively, in both scenarios, a user maintains the same inherent intent and his/her behaviors are likely to be determined by the same demand intent. Therefore, it is feasible to leverage interaction data from both scenarios to reinforce or complement each other’s dual intents for joint intent-aware modeling. Nevertheless, this joint modeling is not trivial due to the following challenges:
幸运的是,在搜索服务中,这些服务通常与同一在线平台上的推荐服务相伴,用户通过查询词明确表达他们的需求意图,如图 1(b)所示。这种明确的搜索需求信息可以作为额外的显式信息,辅助学习推荐中的隐含需求意图。实际上,搜索和推荐任务都旨在理解用户意图,以帮助他们获取所需物品(Belkin 和 Croft, 1992)。此外,在搜索场景中,用户的互动不仅受其显式需求意图影响,还受其个性化固有意图的影响。然而,搜索模型通常侧重于搜索结果与用户需求意图的匹配,往往忽视了个性化固有意图的影响,而这种影响实际上是显著的(Sondhi 等, 2018)。直观上,在这两种场景中,用户保持相同的固有意图,其行为很可能由相同的需求意图决定。因此,利用两种场景的交互数据来相互强化或补充双重意图,进行联合意图感知建模是可行的。然而,这种联合建模并非易事,主要面临以下挑战:
(1) How to accurately model a user’s implicit demand intent in recommendation with search data?
A user’s demand intent is implicit within recommendation but is explicitly indicated by search queries. If the changing demand intents in recommendation can be accurately generated, search and recommendation can be well modeled in a unified manner. The existing method, SRJGraph (Zhao et al., 2022), employs the unchanging padding query in recommendation for unified modeling. This approach assumes an unchanging demand intent across all recommendation interactions, which may hinder recommendation performance. To learn demand intents, an intuitive approach is to simply incorporate users’ historical queries as additional demand information into the recommendation model. However, without explicit supervision to verify the accuracy of demand intents, there may be a significant discrepancy between the learned and the actual demand intents.
(1) 如何在推荐系统中利用搜索数据准确建模用户的隐性需求意图?用户的需求意图在推荐中是隐性的,但在搜索查询中是显性表达的。如果能够准确生成推荐中需求意图的变化,搜索和推荐就能在统一框架下得到良好建模。现有方法 SRJGraph(Zhao 等,2022)在推荐中使用不变的填充查询进行统一建模。这种方法假设所有推荐交互中的需求意图不变,这可能限制推荐性能。为了学习需求意图,一种直观的方法是将用户的历史查询作为额外的需求信息纳入推荐模型。然而,若无明确的监督来验证需求意图的准确性,学习到的需求意图与实际需求意图之间可能存在显著差异。
(2) How to couple the dual intents to model the relation among the intents and the interactive items?
Both inherent intent and demand intent affect the interactive item.
Intuitively, the superimposition of inherent intents (e.g., preferring cheap items) and changing demand intents (needing a T-shirt in summer but a down jacket in winter) leads to changing interactive results (interacting with a cheap T-shirt and cheap down jacket, respectively). In essence, the demand intent can be regarded as the changing deviation from the inherent intent to the changing interactive item.
A common approach is to simply feed the two intents as input features, but it cannot fully capture the relation between the dual intents and the interactive item.
(2) 如何将双重意图耦合以建模意图与交互项之间的关系?内在意图和需求意图均影响交互项。直观上,内在意图(如偏好廉价物品)与变化的需求意图(夏季需要 T 恤而冬季需要羽绒服)的叠加,导致交互结果的变化(分别与廉价 T 恤和廉价羽绒服交互)。本质上,需求意图可视为从内在意图到变化交互项的变动偏差。常见方法是简单地将两种意图作为输入特征,但这无法充分捕捉双重意图与交互项之间的关系。
To tackle these challenges, we propose a novel model named Unified Dual-Intent Translation for joint modeling of Search and Recommendation (UDITSR). Overall, UDITSR comprises a search-supervised demand intent generator and a dual-intent translation module. Specifically, in the demand intent generator, search queries serve as supervision information, allowing us to learn and understand a user’s changing demand intent for recommendations both reliably and accurately.
Moreover, we develop a dual-intent translation propagation mechanism. This mechanism explicitly models the interpretable relation among the triplet elements--user’s ¡inherent intent, demand intent, interactive item¿--within a shared semantic space by employing embedding translations. Particularly, we design an intent translation contrastive learning to further constrain the translation relation. Extensive offline and online experiments were conducted to demonstrate our model’s effectiveness. To gain deeper insights into the effectiveness of our model, we also provide a visual analysis of relevant intents.
为应对这些挑战,我们提出了一种名为统一双意图翻译的联合搜索与推荐模型(UDITSR)。总体而言,UDITSR 包含一个搜索监督的需求意图生成器和一个双意图翻译模块。具体来说,在需求意图生成器中,搜索查询作为监督信息,使我们能够可靠且准确地学习和理解用户对推荐内容不断变化的需求意图。此外,我们开发了一种双意图翻译传播机制。该机制通过嵌入翻译,在共享语义空间中显式建模用户固有意图、需求意图与交互项目三元组元素之间的可解释关系。特别地,我们设计了意图翻译对比学习,以进一步约束翻译关系。通过广泛的线下和线上实验,我们验证了模型的有效性。为深入了解模型的有效性,我们还提供了相关意图的可视化分析。
2. Related Work 2. 相关工作
2.1. Recommendation and Search Models
2.1. 推荐与搜索模型
Recommendation aims to filter items from vast candidate pools to match user interests. Traditional models, such as Collaborative Filtering (CF), assume users with similar behaviors share item preferences (Sarwar et al., 2001; Cheng et al., 2016; He et al., 2017; Guo et al., 2017). Later studies (Zhou et al., 2018; Sun et al., 2019; Feng et al., 2019) focus on decoding users’ evolving interests from their historical behaviors, using techniques like DIN (Zhou et al., 2018), which employs attention mechanism to connect past behaviors with current targets.
Recognizing that users’ interactions are driven by their intrinsic intents, recent studies (Wang et al., 2019b; Chen et al., 2019; Zhu et al., 2020; Wang et al., 2020) exploit users’ historical behavior sequences to understand their changing intents, aiming to better meet user needs. For instance, KA-MemNN (Zhu et al., 2020) uses item categories from user behavior as intent proxies, implementing memory networks for dynamic intent modeling. However, these approaches often deduce intents from interaction behaviors or directly equate behavior with intent, without mining real intrinsic intents. In contrast, our model utilizes the user’s actual demand intents in the search scenario as supervision information to imitate the intents in recommendation.
推荐系统旨在从庞大的候选池中筛选出符合用户兴趣的条目。传统的推荐模型,如协同过滤(CF),假设具有相似行为的用户共享对物品的偏好(Sarwar et al., 2001; Cheng et al., 2016; He et al., 2017; Guo et al., 2017)。后续研究(Zhou et al., 2018; Sun et al., 2019; Feng et al., 2019)则侧重于从用户的历史行为中解码其不断变化的兴趣,采用如 DIN(Zhou et al., 2018)等技术,通过注意力机制将过去行为与当前目标相连接。认识到用户的互动受其内在意图驱动,近期研究(Wang et al., 2019b; Chen et al., 2019; Zhu et al., 2020; Wang et al., 2020)利用用户的历史行为序列来理解其意图的变化,旨在更好地满足用户需求。例如,KA-MemNN(Zhu et al., 2020)使用用户行为中的物品类别作为意图代理,实现记忆网络进行动态意图建模。然而,这些方法往往从交互行为中推断意图,或将行为直接等同于意图,而未深入挖掘真实的内在意图。 相比之下,我们的模型利用用户在搜索场景中的实际需求意图作为监督信息,来模仿推荐中的意图。
Search and recommendation services often coexist on the same platform (Qin et al., 2023). Earlier research (Belkin and Croft, 1992) suggests their goals are essentially equivalent--helping people get the items they want, prompting studies on their joint optimization. For example, JSR (Zamani and Croft, 2018) introduces a shared-parameter framework, with user and item embeddings shared. USER (Yao et al., 2021) treats recommendation behavior as a form of search behavior with unchanging padding query, unifying the modeling of search and recommendation sequences. Furthermore, SRJgraph (Zhao et al., 2022) constructs a unified graph from search and recommendation behaviors, incorporating search queries and a padding query for recommendation as attributes of user-item edges.
These models assume the query-related intents in recommendation are unchanging while the matching degree between the query and the candidate items significantly affects search performance.
This assumption creates a significant gap between the modeling of search and recommendation, greatly hindering the effectiveness of joint modeling approaches. Our model, however, adapts to learn personalized and changing query-related intents for distinct user-item pairs in recommendation, thus enhancing the unification of joint search and recommendation.
搜索与推荐服务常共存于同一平台(Qin 等,2023)。早期研究(Belkin 和 Croft,1992)表明,它们的目标本质上是一致的——帮助人们获取所需物品,这促使了对两者联合优化的研究。例如,JSR(Zamani 和 Croft,2018)引入了共享参数框架,用户与物品嵌入共享。USER(Yao 等,2021)将推荐行为视为一种带有固定填充查询的搜索行为,统一了搜索与推荐序列的建模。此外,SRJgraph(Zhao 等,2022)从搜索与推荐行为构建了一个统一图,将搜索查询及推荐填充查询作为用户-物品边的属性。这些模型假设推荐中的查询相关意图不变,而查询与候选物品的匹配度显著影响搜索性能。这一假设在搜索与推荐的建模之间造成了显著差距,极大阻碍了联合建模方法的有效性。 然而,我们的模型能够适应学习个性化且不断变化的查询相关意图,以区分推荐中的不同用户-项目对,从而增强联合搜索与推荐的一体化。
2.2. Graph Neural Network 2.2. 图神经网络
Graph Neural Networks (GNNs) (Scarselli et al., 2008; Wu et al., 2020) have gained tremendous attention in recent years due to their remarkable ability to process graph-structured data. For instance, Graph Convolutional Network (GCN) (Kipf and Welling, 2016) employs a localized filter to aggregate information from neighbors, and Graph Attention Network (GAT) (Veličković et al., 2018) leverages the attention mechanism to weigh the importance of each neighbor node during the aggregation process. Since then, numerous variants of GNNs (Zhang et al., 2019; Yan et al., 2018; Derr et al., 2018) have been proposed to tackle various types of graphs.
Nowadays, Graph Neural Networks have shown great potential in a wide range of applications, such as recommendation (Wang et al., 2019a; He et al., 2020; Wu et al., 2022) and search (Niu et al., 2020; Fan et al., 2022; Liu et al., 2020a) scenarios. In this work, we propose incorporating demand intents that are generated through search supervision in recommendation scenario, as well as explicitly stated search intents, into the construction of a unified graph. Specifically, these demand intents serve as the attributes of the edges connecting users and items. Moreover, the invariant node representations for a user across different interactions are used to indicate their inherent intents. Based on the graph, we propose a novel dual-intent translation propagation for unified dual intent-aware modeling.
图神经网络(GNNs)(Scarselli 等,2008;Wu 等,2020)近年来因其处理图结构数据的卓越能力而备受瞩目。例如,图卷积网络(GCN)(Kipf 和 Welling,2016)采用局部滤波器从邻居节点聚合信息,而图注意力网络(GAT)(Veličković等,2018)则利用注意力机制在聚合过程中为每个邻居节点赋予重要性权重。自此,众多 GNN 变体(Zhang 等,2019;Yan 等,2018;Derr 等,2018)被提出,以应对各类图结构。如今,图神经网络在推荐(Wang 等,2019a;He 等,2020;Wu 等,2022)和搜索(Niu 等,2020;Fan 等,2022;Liu 等,2020a)等广泛应用场景中展现出巨大潜力。在本研究中,我们提出将通过搜索监督生成的推荐场景需求意图,以及明确表达的搜索意图,融入到统一图的构建中。 具体而言,这些需求意图作为连接用户与物品的边的属性。此外,用户在不同交互中的不变节点表示用于指示其固有意图。基于该图,我们提出了一种新颖的双意图翻译传播方法,用于统一的双意图感知建模。
3. Preliminary 3. 初步
Let and denote the universal sets of users and items in both search and recommendation scenarios. In order to distinguish these two scenarios, we define the interaction records in each scenario as follows:
设 和 分别表示搜索和推荐场景中用户和物品的全集。为了区分这两种场景,我们将各场景中的交互记录定义如下:
Definition 1.search scenario: In the search data , each interaction record can be formulated as , which represents that user clicked item with the explicit query . The query can
be segmented into several shorter terms as , where denotes the -th term and is the number of terms in query .
定义 1.搜索场景:在搜索数据 中,每个交互记录 可以表述为 ,表示用户 通过显式查询 点击了项目 。查询 可以被分割成若干较短的词项,如 ,其中 表示第 个词项, 是查询 中的词项总数。
Definition 2.recommendation scenario: In the recommendation data , each interaction record can be formulated as , which represents user clicked item without an explicit query.
定义 2.推荐场景:在推荐数据 中,每个交互记录 可以表示为 ,这代表用户 在没有明确查询的情况下点击了项目 。
Thereby, the double-scenario graph including all user click behaviors in both scenarios can be constructed as follows:
由此,包含两个场景中所有用户点击行为的双场景图可以构建如下:
Definition 3.double-scenario graph: Given the set of all user click behaviors in both scenarios, denoted as , the double-scenario graph can be formulated as . Each search edge corresponds to a record () in , while
each recommendation edge corresponds to a record () in .
定义 3.双场景图:给定两种场景下所有用户点击行为的集合,记作 ,双场景图可表示为 。每条搜索边 对应于 中的一条记录( ),而每条推荐边 则对应于 中的一条记录( )。
In Figure 2(a), there is an example of our double-scenario graph. For instance, user searches for query and then clicks item in search scenario. Thus, an edge exists between nodes and , with query assigned as an attribute of this edge. Likewise, in recommendation scenario, when user clicks item , an edge also exists between user and item , but without any query attribute. Based on the above definitions, the joint modeling of search and recommendation can be defined as follows:
在图 2(a)中,展示了一个双场景图的示例。例如,用户 搜索了查询 ,随后在搜索场景中点击了项目 。因此,节点 和 之间存在一条边,查询 被指定为该边的属性。同样地,在推荐场景中,当用户 点击项目 时,用户 和项目 之间也存在一条边,但没有附加任何查询属性。基于上述定义,搜索与推荐的联合建模可定义如下:
Problem definition.
Given search data , recommendation data and double-scenario graph , this task is to train a joint model of search and recommendation to predict the most appropriate items that user will interact.
问题定义。给定搜索数据 、推荐数据 及双场景图 ,本任务旨在训练一个搜索与推荐联合模型,以预测用户 最可能互动的合适项目 。
4. Methodology 4. 方法论
In this section, we introduce UDITSR for dual intent-aware joint modeling of search and recommendation, as depicted in Figure 2. We begin with the model’s embedding layer in Section 4.1. Then, in Section 4.2, we detail a search-supervised demand intent generator that leverages search query data to infer recommendation intents, which allows us to convert the double-scenario graph into a unified graph. Utilizing this graph, we describe dual-intent translation propagation to couple inherent intents and demand intents, enhanced by a contrastive loss to constrain the translation relation. Finally, the prediction layer and optimization are illustrated in Section 4.4.
在本节中,我们介绍 UDITSR,用于搜索与推荐的双重意图感知联合建模,如图 2 所示。首先,我们在第 4.1 节中介绍模型的嵌入层。接着,在第 4.2 节中,我们详细阐述了一个搜索监督的需求意图生成器,该生成器利用搜索查询数据推断推荐意图,从而使我们能够将双重场景图转换为统一图。利用此图,我们描述了双重意图的翻译传播,通过对比损失来约束翻译关系,以耦合固有意图和需求意图。最后,预测层和优化过程在第 4.4 节中展示。
4.1. Embedding Layer 4.1. 嵌入层
By feeding the user ID and item ID into the user and item embedding matrices respectively, we can obtain the embeddings of user and item as . Since each query is a sequence of shorter terms as [, , , ], we can obtain the representation of query by combining the embeddings of its terms:
通过将用户 ID 和物品 ID 分别输入用户和物品嵌入矩阵,我们可以获得用户 和物品 的嵌入表示为 。由于每个查询 是由较短的词项序列[ , , , ]组成,我们可以通过组合这些词项的嵌入来获得查询 的表示:
(1) |
where represents the embedding of the -th query term in and denotes a combination function. In this study, we choose the element-wise sum-pooling operation because it is both efficient and effective for this combination through empirical analysis.
其中 表示 中第 个查询词的嵌入, 表示组合函数。在本研究中,我们选择逐元素求和池化操作,因为通过实证分析发现,它在组合过程中既高效又有效。
4.2. Demand Intent Generation
4.2. 需求意图生成
4.2.1. Search-Supervised Demand Intent Generator
4.2.1. 搜索监督的需求意图生成器
The notable difference between search and recommendation is that a user explicitly expresses demand intents in search, whereas recommendation lacks such explicit intents. To bridge this gap, we propose to utilize the abundant query information from search to supervise the generation of users’ demand intents in recommendation. Below we describe the generator in detail.
搜索与推荐之间显著的区别在于,用户在搜索中明确表达了需求意图,而推荐则缺乏这种明确的意图。为了弥合这一差距,我们提出利用搜索中丰富的查询信息来监督推荐中用户需求意图的生成。以下详细描述生成器。
Since the user’s historical queries and the item’s historical queries contain abundant demand intent information, we leverage them as auxiliary information to simulate the user’s demand intent for recommendation. Similar to the processing of in Eq. 1, we adopt the element-wise sum-pooling operation to obtain the representation of as , where is the embedding of the -th query term in . Since contains query words from multiple users, we introduce a user-aware gate mechanism to model personalized demand intents.
Particularly, the user-aware gating network yields a distribution over the query words. The personalized representation of is then formulated as the weighted sum of the embeddings of its query words, as follows:
由于用户的历史查询 和物品的历史查询 蕴含丰富的需求意图信息,我们将其作为辅助信息来模拟用户的需求意图以进行推荐。类似于在公式 1 中处理 的方式,我们采用逐元素求和池化操作来获取 的表示为 ,其中 是 中第 个查询词的嵌入。由于 包含来自多个用户的查询词,我们引入了一种用户感知的门控机制来建模个性化的需求意图。特别地,用户感知的门控网络 生成一个关于 查询词的分布。 的个性化表示则被表述为其查询词嵌入的加权和,如下所示:
(2) | ||||
where denotes the concatenation operation; is used to match the dimensions of vector and the concatenated vector. Then, with user-related representations and item-related representations , the user’s demand intent about the item can be estimated as follows:
其中 表示连接操作; 用于匹配向量 与连接向量的维度。随后,利用用户相关表示 和物品相关表示 ,可以估计用户对物品的需求意图如下:
(3) |
where denotes a multi-layer perceptron. Since the ground truth queries in search data serve as the supervision information for generating demand intent, we design the generation loss as follows:
其中 表示多层感知器。由于搜索数据中的真实查询作为生成需求意图的监督信息,我们将生成损失设计如下:
(4) |
4.2.2. Unified Graph 4.2.2. 统一图
After generating the demand intents, each recommendation record () in can be converted into a triplet (), where the embedding of corresponds to the generated intents . For simplicity, we directly generate the representation of intent instead of indirectly predicting the specific query . With the generated demand intents, the double-scenario graph can be converted into a unified graph.
Specifically, an additional attribute is attached to each recommendation edge () in . For brevity, we use to uniformly represent the real in search scenario and the generated in recommendation scenario correspondingly. Based on the unified graph, we implement the unified modeling of recommendation and search below.
在生成需求意图后, 中的每条推荐记录( )可以转化为一个三元组( ),其中 的嵌入对应于生成的意图 。为简化起见,我们直接生成意图 的表示,而非间接预测具体的查询 。借助生成的需求意图,双场景图可以转化为统一图。具体来说,在 中的每条推荐边( )上附加了一个额外属性 。为简洁起见,我们用 统一表示搜索场景中的实际 和推荐场景中生成的 。基于统一图,我们在下方实现了推荐与搜索的统一建模。
4.3. Dual-Intent Translation Propagation
4.3. 双意图翻译传播
To explicitly model the relation among the dual intents and the interactive items, we propose a dual-intent translation module inspired by the triplet-based representation learning in knowledge graphs (Bordes et al., 2013). Specifically, we use the user’s embedding representation, which remains inherent for a single user, to represent their inherent intent. The search query representation and the generated demand intent in recommendation represent the user’s demand intent. The representation of an interactive item is given by its embedding. We assume that a user’s changing interactive item should be close to their inherent intent plus changing demand intent.
Consequently, we aggregate the neighbor embeddings as follows:
为了显式建模双意图与交互项之间的关系,我们提出了一种受知识图谱中基于三元组的表示学习启发的双意图翻译模块(Bordes 等,2013)。具体而言,我们使用用户的嵌入表示,该表示对单个用户保持固有特性,来代表其固有意图。搜索查询表示和推荐中生成的需求意图共同代表了用户的需求意图。交互项的表示则由其嵌入给出。我们假设,用户交互项的变化应接近其固有意图加上变化的需求意图。因此,我们按如下方式聚合邻居嵌入:
(5) | ||||
where and denote the neighboring nodes of user and item respectively, in the unified graph; and .
In particular, the subtraction aggregation operation, as opposed to the addition operation, for aggregating the embeddings of user neighboring nodes to simulate users’ inherent intents.
Finally, the weighted-pooling operation is applied to generate the aggregated representations by operating on the propagated layers:
其中, 和 分别表示用户 和物品 在统一图中的相邻节点; 和 。特别地,为了模拟用户固有的意图,采用了减法聚合操作,而非加法操作,来聚合用户相邻节点的嵌入。最后,通过在传播的 层上进行加权池化操作,生成聚合表示:
(6) |
where indicates the importance of the -th layer representation in constituting the final embedding. Following LightGCN (He et al., 2020), we set as , as the focus of our work is not on its selection.
其中 表示在构成最终嵌入时,第 层表示的重要性。遵循 LightGCN(He 等,2020)的做法,我们将 设为 ,因为我们的工作重点不在于此选择。
To further constrain the translation relation, we design an intent translation contrastive learning approach that adopts a margin-based ranking criterion. Specifically, we aim to ensure that (i.e., the ground truth interactive item should be near to the translated intent ), while the negative should be distant from , as follows:
为了进一步约束翻译关系,我们设计了一种意图翻译对比学习方法,采用基于边界的排序准则。具体而言,我们的目标是确保 (即真实交互项 应接近翻译意图 ),同时负样本 应远离 ,如下所示:
(7) |
where denotes the representation of real query in search or the generated demand intent in recommendation for () pair; denotes the pairwise training data where indicates the positive observed interaction set, and represents the randomly-sampled negative set; stands for the sigmoid function.
其中 表示搜索中的真实查询或推荐中的生成需求意图; 表示( )对; 表示成对训练数据,其中 表示正向观察到的交互集合, 表示随机采样的负样本集合; 代表 Sigmoid 函数。
4.4. Model Prediction and Optimization
4.4.模型预测与优化
After obtaining the representations , we fuse them to obtain the overall representation for the input sample :
在获得表示 后,我们将它们融合以获得输入样本 的整体表示:
(8) |
Then, two different MLPs are employed to make prediction for search and recommendation tasks, respectively:
然后,分别采用两个不同的多层感知机(MLP)来对搜索和推荐任务进行预测:
(9) |
We adopt pairwise training to train the model.
Specifically, we adopt the Bayesian Personalized Ranking (BPR) (Rendle et al., 2009) loss to emphasize that the observed interaction should be assigned a higher score than the unobserved one as follows:
我们采用成对训练来训练模型。具体而言,我们采用贝叶斯个性化排序(BPR)(Rendle 等人,2009 年)损失,强调观察到的交互应被赋予比未观察到的交互更高的分数,如下所示:
(10) |
where the representation of denotes the demand intent for the negative pair (). Finally, the overall loss is defined using hyper-parameters and as:
其中, 表示负样本对 ( ) 的需求意图。最后,整体损失 使用超参数 和 定义为:
(11) |
5. Experiments 5.实验
In this section, we present empirical results to demonstrate the effectiveness of our proposed UDITSR. These experiments are designed to answer the following research questions: RQ1 How does UDITSR perform compared with state-of-the-art search and recommendation models?
RQ2 What are the effects of the demand intent generator and dual-intent translation mechanism in UDITSR?
RQ3 Why could UDITSR perform better?
RQ4 How does UDITSR perform in real-world online recommendations with practical metrics?
RQ5 How do the hyper-parameters in UDITSR impact the search and recommendation performance?
在本节中,我们展示实证结果以证明我们提出的 UDITSR 的有效性。这些实验旨在回答以下研究问题:RQ1 UDITSR 与最先进的搜索和推荐模型相比表现如何?RQ2 需求意图生成器和双意图翻译机制在 UDITSR 中的效果如何?RQ3 为何 UDITSR 能表现更佳?RQ4 UDITSR 在实际在线推荐中,以实际指标衡量表现如何?RQ5 UDITSR 中的超参数如何影响搜索和推荐性能?
5.1. Experimental Settings
5.1. 实验设置
5.1.1. Dataset Description
5.1.1. 数据集描述
We conducted experiments on two real-world datasets, denoted as MT-Large and MT-Small datasets111We collected this dataset because there was no public dataset that includes both search and recommendation data. Our code and data will be available at https://github.com/17231087/UDITSR.. These two datasets are obtained from the Meituan platform, one of the largest takeaway platforms in China. Both datasets span eight days across two cities.
Each sample in the datasets contains a user and an item, and each search sample additionally contains a query. Specifically, with 111,891 search and 65,035 recommendation interactions collected, our MT-Small dataset comprises 56,887 users and 4,059 items and the average number of split words per query record is 1.6801. With 1,527,869 search and 1,168,491 recommendation interactions collected, the MT-Large dataset contains 433,573 users and 22,967 items and the average number of split words per query is 1.5561.
To evaluate model performance, we split the first six days’ data for training, the seventh day’s data for validation, and the last day’s data for testing.
For each ground truth test record, we randomly sampled 99 items that the user did not interact with as negative samples.
我们在两个真实世界的数据集上进行了实验,分别标记为 MT-Large 和 MT-Small 数据集。这两个数据集来自中国最大的外卖平台之一——美团平台。两个数据集涵盖了两个城市的八天时间。每个样本包含一个用户和一个物品,每个搜索样本还额外包含一个查询。具体而言,MT-Small 数据集收集了 111,891 次搜索和 65,035 次推荐互动,包含 56,887 名用户和 4,059 个物品,每条查询记录的平均分词数为 1.6801。MT-Large 数据集则收集了 1,527,869 次搜索和 1,168,491 次推荐互动,包含 433,573 名用户和 22,967 个物品,每条查询的平均分词数为 1.5561。为了评估模型性能,我们将前六天的数据用于训练,第七天的数据用于验证,最后一天的数据用于测试。对于每个真实测试记录,我们随机抽取了 99 个用户未互动的物品作为负样本。
表 1.网络配置
Name 姓名 | Value 价值 |
optimizer 优化器 | AdamW 亚当 W |
batch size 批量大小 | 256 |
learning rate 学习率 | 1e-4 |
weight decay 权重衰减 | 1e-5 |
vocab size of words in querys 查询中词语的词汇量 |
5,000 |
dimension of embeddings 嵌入维度 | 100 |
depth of aggregation 聚合深度 | 2 |
number of words per query 每条查询的词数 |
3 |
number of words per user’s historical query 用户历史查询的词数 |
3 |
number of words per item’s historical query 每个商品历史查询的词数 |
10 |
hidden sizes of in demand intent generator 需求意图生成器中 的隐藏尺寸 |
[200,100] |
hidden sizes of / / 的隐藏尺寸 |
[150,75] |
5.1.2. Implementation Details
5.1.2. 实施细节
We implement all models using PyTorch222https://pytorch.org/, a well-known software library for deep learning. In Section 5.6, we report the impact of essential hyper-parameters in our model, including the loss weights and , and we utilize the best settings for these hyper-parameters. The remaining network configurations are presented in Table 1.
To ensure a fair comparison, we apply the above-mentioned settings across all models. Moreover, we search for optimal values of the other hyper-parameters of the baseline models as suggested in their respective original papers. Finally, we employ the early stopping strategy based on the models’ performance on the validation set to avoid overfitting.
我们使用 PyTorch 2 实现所有模型,PyTorch 是一个著名的深度学习软件库。在第 5.6 节中,我们报告了模型中关键超参数的影响,包括损失权重 和 ,并采用了这些超参数的最佳设置。其余的网络配置在表 1 中展示。为确保公平比较,我们将上述设置应用于所有模型。此外,我们根据各自原始论文的建议,搜索基线模型其他超参数的最优值。最后,我们采用基于验证集上模型性能的早停策略,以防止过拟合。
Dataset 数据集 | Model | Search 搜索 | Recommendation 推荐 | ||||||
---|---|---|---|---|---|---|---|---|---|
Hit@5 命中@5 | NDCG@5 | MRR | Avg.C 平均 C | Hit@5 命中@5 | NDCG@5 | MRR | AUC | ||
MT-Small 小型机器翻译 | NeuMF 神经矩阵分解 | 0.5510 | 0.4264 | 0.4150 | 12.1907 | 0.3147 | 0.2306 | 0.2374 | 0.8160 |
DNN | 0.5877 | 0.4594 | 0.4465 | 9.6208 | 0.3241 | 0.2177 | 0.2246 | 0.8150 | |
xDeepFM 深度交叉因子分解机 | 0.5053 | 0.3886 | 0.3815 | 14.3603 | 0.3184 | 0.2139 | 0.2218 | 0.8155 | |
DIN | 0.5892 | 0.4726 | 0.4613 | 11.6023 | 0.3632 | 0.2510 | 0.2545 | 0.8213 | |
AEM | 0.5053 | 0.3666 | 0.3568 | 11.7953 | 0.3967 | 0.2703 | 0.2686 | 0.7982 | |
TEM | 0.5362 | 0.4185 | 0.4084 | 13.6472 | 0.2933 | 0.1970 | 0.2078 | 0.7947 | |
JSR | 0.6143 | 0.4828 | 0.4678 | 8.6276 | 0.3460 | 0.2448 | 0.2457 | 0.7532 | |
SimpleX 简单 X | 0.6237 | 0.4864 | 0.4699 | 8.0841 | 0.3314 | 0.2288 | 0.2336 | 0.8081 | |
MGDSPR | 0.6150 | 0.4743 | 0.4570 | 8.5362 | 0.2974 | 0.2032 | 0.2122 | 0.7862 | |
GAT | 0.6025 | 0.4679 | 0.4497 | 10.8707 | 0.4202 | 0.3109 | 0.3032 | 0.7935 | |
NGCF | 0.6418 | 0.5173 | 0.5000 | 9.7943 | 0.4564 | 0.3346 | 0.3284 | 0.8232 | |
LightGCN 轻量图卷积网络 | 0.6665 | 0.5402 | 0.5195 | 9.9139 | 0.4577 | 0.3296 | 0.3185 | 0.8174 | |
GraphSRRL 图谱 SRRL | 0.6688 | 0.5267 | 0.5042 | 8.0724 | 0.4540 | 0.3249 | 0.3159 | 0.7883 | |
SRJgraph SRJ 图表 | 0.6186 | 0.4850 | 0.4647 | 12.1989 | 0.4140 | 0.3074 | 0.2997 | 0.7474 | |
DCCF | 0.5013 | 0.3760 | 0.3615 | 19.6477 | 0.4323 | 0.3380 | 0.3304 | 0.7239 | |
UDITSR | 0.7008* | 0.5691* | 0.5470* | 7.5257* | 0.4841* | 0.3528* | 0.3422* | 0.8285* | |
Impr.% 印量% | 4.7847 | 5.3499 | 5.2936 | 6.7725 | 5.7680 | 4.3787 | 3.5714 | 0.6438 | |
MT-Large MT-大型 | NeuMF 神经矩阵分解 | 0.8668 | 0.7855 | 0.7682 | 3.3001 | 0.5390 | 0.4235 | 0.4129 | 0.8573 |
DNN | 0.8788 | 0.7874 | 0.7664 | 3.0520 | 0.5153 | 0.3962 | 0.3881 | 0.8610 | |
xDeepFM 深度交叉因子分解机 | 0.8552 | 0.7417 | 0.7147 | 4.0106 | 0.4926 | 0.3897 | 0.3828 | 0.8077 | |
DIN | 0.8914 | 0.7934 | 0.7693 | 2.7283 | 0.6005 | 0.4489 | 0.4292 | 0.9082 | |
AEM | 0.8760 | 0.7654 | 0.7389 | 2.9007 | 0.5865 | 0.4597 | 0.4435 | 0.8940 | |
TEM | 0.8611 | 0.7522 | 0.7269 | 3.4096 | 0.5031 | 0.3526 | 0.3419 | 0.8899 | |
JSR | 0.8691 | 0.7748 | 0.7537 | 3.0903 | 0.5023 | 0.3789 | 0.3683 | 0.8393 | |
SimpleX 简单 X | 0.8896 | 0.7895 | 0.7651 | 2.6466 | 0.5004 | 0.3790 | 0.3691 | 0.8640 | |
MGDSPR | 0.8726 | 0.7709 | 0.7473 | 3.0325 | 0.5412 | 0.4037 | 0.3888 | 0.8751 | |
GAT | 0.8761 | 0.7796 | 0.7572 | 2.9530 | 0.5880 | 0.4540 | 0.4347 | 0.8706 | |
NGCF | 0.8821 | 0.7892 | 0.7670 | 2.7377 | 0.6325 | 0.4966 | 0.4780 | 0.9096 | |
LightGCN 轻量图卷积网络 | 0.8937 | 0.8016 | 0.7795 | 2.4076 | 0.6158 | 0.4785 | 0.4593 | 0.8920 | |
GraphSRRL 图谱 SRRL | 0.8966 | 0.7992 | 0.7755 | 2.3504 | 0.6106 | 0.4726 | 0.4543 | 0.8891 | |
SRJgraph SRJ 图表 | 0.8836 | 0.7883 | 0.7659 | 2.6849 | 0.5873 | 0.4494 | 0.4315 | 0.8942 | |
DCCF | 0.8568 | 0.7459 | 0.7205 | 3.2666 | 0.6201 | 0.5007 | 0.4802 | 0.8292 | |
UDITSR | 0.9178* | 0.8382* | 0.8183* | 1.9819* | 0.6566* | 0.5157* | 0.4936* | 0.9146* | |
Impr.% 印量% | 2.3645 | 4.5659 | 4.9775 | 15.6782 | 3.8103 | 2.9958 | 2.7905 | 0.5497 |
表 2. 两个数据集上的总体表现。 表示 Avg.C 指标值越小,性能越好。Impr.%表示最佳表现方法(加粗)相对于最强基线(下划线)的相对改进。*表示通过配对 t 检验,UDITSR 与最佳基线相比,在 0.05 显著性水平上的差异。
5.1.3. Evaluation Metrics 5.1.3. 评估指标
To evaluate our model’s performance, we utilize four widely-used ranking metrics: Hit@K, NDCG@K (Järvelin and Kekäläinen, 2002) (we set K as 5 by default), MRR (Radev et al., 2002) and Average position of the Clicked items (Avg.C) (Yao et al., 2021). Additionally, we adopt an accuracy metric, AUC (Ferri et al., 2011) for the recommendation task.
为了评估我们模型的性能,我们采用了四种广泛使用的排序指标:Hit@K、NDCG@K(Järvelin 和 Kekäläinen,2002)(默认设置 K 为 5)、MRR(Radev 等人,2002)以及点击项目的平均位置(Avg.C)(Yao 等人,2021)。此外,我们还采用了推荐任务的准确性指标 AUC(Ferri 等人,2011)。
5.1.4. Baselines 5.1.4. 基线
In our work, we evaluate the performance of our model with two groups of baselines to examine its effectiveness.
在我们的工作中,我们通过两组基线评估了模型的性能,以检验其有效性。
(1) Graph-free baselines (1)无图基线
-
•
NeuMF (He et al., 2017) combines traditional matrix decomposition with the MLP to extract low-dimensional and high-dimensional features simultaneously.
NeuMF(He 等,2017)将传统的矩阵分解与多层感知机(MLP)相结合,以同时提取低维和高维特征。 - •
-
•
xDeepFM (Lian et al., 2018) consists of a compressed interaction network (CIN) and an MLP for prediction, where CIN generates explicit feature interactions at the vector-wise level.
xDeepFM(Lian 等,2018)由一个压缩交互网络(CIN)和一个用于预测的多层感知机(MLP)组成,其中 CIN 在向量级别上生成显式特征交互。 -
•
DIN (Zhou et al., 2018) utilizes an attention mechanism between the historical behavior sequence and the target item to model the evolving interests.
DIN(周等人,2018)利用历史行为序列与目标商品之间的注意力机制,来建模不断演变的兴趣。 -
•
AEM (Ai et al., 2019) allocates different attention values to the previous behavior sequence based on the current search queries.
AEM(Ai 等,2019)根据当前搜索查询,为先前的行为序列分配不同的注意力值。 -
•
TEM (Bi et al., 2020) feeds the sequence of query and user behavior history into a transformer layer to extract the search intents.
TEM(Bi 等,2020)将查询序列和用户行为历史输入到 Transformer 层中,以提取搜索意图。 -
•
JSR (Zamani and Croft, 2020) integrates neural collaborative filtering and language modeling to reconstruct query text descriptions, enabling the joint model of search and recommendation.
JSR(Zamani 和 Croft,2020)将神经协同过滤与语言建模相结合,以重构查询文本描述,从而实现搜索与推荐的联合模型。 -
•
SimpleX (Mao et al., 2021) is a simplified variant of the two-tower model with user behavior modeling.
SimpleX(Mao 等,2021)是双塔模型的一个简化变体,专注于用户行为建模。 -
•
MGDSPR (Li et al., 2021) utilizes an attention mechanism to model the relationship between users’ query multi-grained semantics and their personalized behaviors for prediction.
MGDSPR(Li 等,2021)利用注意力机制来建模用户查询的多粒度语义与其个性化行为之间的关系,以进行预测。
(2) Graph-based baselines
(2) 基于图的基线方法
-
•
GAT (Veličković et al., 2018) utilizes the attention mechanism to measure the importance of neighbor nodes during the aggregation process.
GAT(Veličković等,2018)利用注意力机制在聚合过程中衡量邻居节点的重要性。 -
•
NGCF (Wang et al., 2019a) enhances the Graph Convolutional Networks (GCN) by incorporating user-item interactions.
NGCF(Wang 等,2019a)通过融入用户-物品交互,增强了图卷积网络(GCN)。 -
•
LightGCN (He et al., 2020) streamlines GCN by relying solely on neighborhood aggregation to capture collaborative filtering, omitting feature transformation and non-linear activation components.
LightGCN(He 等,2020)通过仅依赖邻域聚合来捕捉协同过滤,简化了 GCN,省略了特征变换和非线性激活组件。 -
•
GraphSRRL (Liu et al., 2020a) exploits three specific structural patterns within a user-query-item graph.
GraphSRRL(Liu 等,2020a)利用了用户-查询-项目图中的三种特定结构模式。 -
•
SRJGraph (Zhao et al., 2022) incorporates padding queries for recommendation and search queries as attributes into interaction edges, enabling joint modeling of both tasks.
SRJGraph(赵等人,2022)将推荐和搜索查询的填充查询作为属性融入交互边中,实现了两项任务的联合建模。 -
•
DCCF (Ren et al., 2023) leverages an adaptive self-supervised augmentation to disentangle intents behind user-item interactions.
DCCF(Ren 等人,2023)利用自适应自监督增强技术来解构用户-物品交互背后的意图。
Specifically, NeuMF, xDeepFM, DIN, DCCF, SimpleX, NGCF and LightGCN are proposed for the recommendation task, while AEM, TEM, MGDSPR and GraphSRRL are proposed for the search task. JSR and SRJGraph are designed for joint learning of both tasks. To adapt these baselines for both tasks, real query representations for search and padding query representations for recommendation are incorporated into the prediction layer described in Section 4.4. Previous studies (Zamani and Croft, 2018; Zhao et al., 2022) have demonstrated that joint optimization of search and recommendation models can improve performance, so all baselines are directly trained on both search and recommendation data.
All baselines use the same settings for the embedding layer and the prediction layer, and the interaction graph is built on both search and recommendation interactions.
具体而言,NeuMF、xDeepFM、DIN、DCCF、SimpleX、NGCF 和 LightGCN 被提出用于推荐任务,而 AEM、TEM、MGDSPR 和 GraphSRRL 则针对搜索任务提出。JSR 和 SRJGraph 设计用于两项任务的联合学习。为使这些基线适应两项任务,在第 4.4 节描述的预测层中,引入了搜索的真实查询表示和推荐的填充查询表示。先前研究(Zamani 和 Croft, 2018; Zhao 等, 2022)表明,搜索与推荐模型的联合优化能提升性能,因此所有基线均直接在搜索和推荐数据上进行训练。所有基线在嵌入层和预测层的设置相同,且交互图基于搜索和推荐交互构建。
5.2. Overall Performance Comparison (RQ1)
5.2. 整体性能比较(RQ1)
We present the results on the two adopted datasets in Table 2. From the results, we can observe that:
我们在表 2 中展示了两个采用数据集的结果。从结果中,我们可以观察到:
-
•
UDITSR significantly outperforms all the competitive baselines on both tasks. Specifically, compared to the best-performing baselines, UDITSR gains an average improvement of 6.22% and 3.06% in the search and recommendation tasks, respectively.
UDITSR 在两项任务中均显著优于所有竞争基线。具体而言,与表现最佳的基线相比,UDITSR 在搜索和推荐任务中分别实现了 6.22%和 3.06%的平均提升。 -
•
Most graph-based methods, such as NGCF, LightGCN, and GraphSRRL, perform well in both tasks, potentially due to their ability to effectively capture complex high-order interactive patterns.
大多数基于图的方法,如 NGCF、LightGCN 和 GraphSRRL,在两项任务中表现出色,这可能归功于它们有效捕捉复杂高阶交互模式的能力。 -
•
SRJgraph assumes that query-related intents in recommendation remain unchanging whereas in search, the matching degree between the query and the candidate items is deemed crucial. Consequently, such an assumption may limit the model’s performance, particularly when compared to our UDITSR, which learns and adapts to changing query-related intents.
SRJgraph 假设推荐中的查询相关意图保持不变,而在搜索中,查询与候选项目之间的匹配程度被视为至关重要。因此,这种假设可能会限制模型的性能,尤其是与我们的 UDITSR 相比,后者能够学习和适应不断变化的查询相关意图。
5.3. Ablation Study (RQ2) 5.3. 消融研究(RQ2)
As the demand intent generator and dual-intent translation propagation are the core of our model, we conduct the following ablation studies to investigate their effectiveness:
由于需求意图生成器和双意图翻译传播是我们模型的核心,我们进行了以下消融研究以探究其有效性:
-
•
UDITSR(w/o DeIntGen) masks all generated demand intents by assigning the embedding of the padding query to each recommended record.
-
•
UDITSR(w/o IntTrans) replaces dual-intent translation with classical mean-pooling propagation between the user and item nodes.
-
•
UDITSR(w/o DeIntGen & IntTrans) removes both the demand intent generator and dual-intent translation propagation, as described in the two ablation studies above.
Dataset | Ablation | Search | Recommendation | ||||||
---|---|---|---|---|---|---|---|---|---|
Hit@5 | NDCG@5 | MRR | Avg.C | Hit@5 | NDCG@5 | MRR | AUC | ||
MT-Small | UDITSR(w/o DeIntGen & IntTrans) | 0.6479 | 0.5152 | 0.4949 | 10.1960 | 0.4185 | 0.3052 | 0.3032 | 0.8179 |
UDITSR(w/o IntTrans) | 0.6454 | 0.5130 | 0.4924 | 10.5527 | 0.4352 | 0.3206 | 0.3154 | 0.8225 | |
UDITSR(w/o DeIntGen) | 0.6959 | 0.5543 | 0.5307 | 8.1389 | 0.4510 | 0.3237 | 0.3178 | 0.8186 | |
UDITSR | 0.7008* | 0.5691* | 0.5470* | 7.5257* | 0.4841* | 0.3528* | 0.3422* | 0.8285* | |
MT-Large | UDITSR(w/o DeIntGen & IntTrans) | 0.8660 | 0.7586 | 0.7337 | 3.1185 | 0.6183 | 0.4818 | 0.4623 | 0.9031 |
UDITSR(w/o IntTrans) | 0.8870 | 0.7866 | 0.7623 | 2.6399 | 0.6228 | 0.4879 | 0.4696 | 0.9061 | |
UDITSR(w/o DeIntGen) | 0.9089 | 0.8192 | 0.7969 | 2.2053 | 0.6303 | 0.4890 | 0.4685 | 0.9058 | |
UDITSR | 0.9178* | 0.8382* | 0.8183* | 1.9819* | 0.6566* | 0.5157* | 0.4936* | 0.9146* |
From the results of ablation studies in Table 3, we can find that:
-
•
UDITSR(w/o DeIntGen & IntTrans) performs the worst on both search and recommendation tasks, suggesting that the significant improvement of our model stems from our proposed demand intent generator and dual-intent translation propagation.
-
•
UDITSR(w/o IntTrans) performs worse than the original UDITSR, highlighting the effectiveness of our proposed intent translation propagation mechanism.
-
•
UDITSR(w/o DeIntGen) performs worse than UDITSR, especially for recommendation task, indicating that the search-supervised demand intent generator can help UDITSR learn implicit intents more accurately in recommendation.
5.4. Intent Visualization (RQ3)
In this section, we visualize the learned intents to further investigate why our model performs better. We compare UDITSR with its ablated version without the dual-intent translation propagation (UDITSR(w/o IntTrans), detailed in Section 5.3). We employ the default setting of the t-SNE (Donahue et al., 2014) provided by Scikit-learn to visualize the distribution of the learned intents and the interactive items. For clarity, we randomly sample 100 positive records from the search and recommendation test datasets respectively for plotting. Specifically, in UDITSR(w/o IntTrans), user embeddings (i.e.,) are regarded as the learned intents, as shown in Figures 3(a) and (d), similar to preference/intent captured by models like NGCF and LightGCN. UDITSR, however, couples inherent and demand intents via intent translation to form the final intents (i.e., ), as shown in Figure 3(c) and (f). To ensure a fair comparison, we present the inherent intents (i.e., ) learned by UDITSR in Figure 3(b) and (e).
Ideally, the distribution of learned intents should match that of interactive item representations. Figures 3(a) and (d) reveal that the intents learned by UDITSR(w/o IntTrans) are concentrated while the positive interactive items are scattered, indicating a mismatch. Meanwhile, the inherent intents learned by UDITSR are relatively scattered, indicating that our model can better learn the personalized inherent intents of different users. However, there still exist obvious gaps between the intents and items, highlighting the necessity of learning demand intents. In contrast, the translated intents learned by UDITSR are scattered in the space of the target interactive items, demonstrating its excellent intent modeling capability. The better fit of the distribution of translated intents to the target interactive distribution could be the fundamental reason for the better overall performance of UDITSR.
5.5. Online A/B test (RQ4)
Owing to the distinct architectural differences between the search and recommender systems on the Meituan Waimai platform, we have initially focused our methodological deployment on the homepage recommender systems. We conducted a month-long online A/B test from December 18, 2023, to January 17, 2024. Specifically, we utilized the search data with query information to guide the learning of user demand intent representation and leveraged the learned graph embeddings as additional features in the downstream recommendation model. The control bucket was the original online recommendation method of Meituan Waimai platform. The deployment of our method increased the GMV(Gross Merchandise Volume) by 1.46% and CTR(Click-Through Rate) by 0.77%, which demonstrated the effectiveness of our method. In the future, we will continue to conduct comprehensive online experiments that encompass both search and recommendation scenarios.
5.6. Hyper-Parameter Studies (RQ5)
In this section, we conduct experiments on the loss weights (, ) in Eq. 11 on MT-Small dataset to explore their impact.
(1) Loss weight of the demand intent generator (). We vary within . The results in Figure 4 indicate that performance improves and then declines with increasing . With , the demand intent generator degenerates to an ordinary generator without any search-supervision information. All models with search supervision (i.e. ) outperform models without it (i.e. ). This may stem from UDITSR’s effective learning of user demand intents through explicit supervision from search. Furthermore, our model excels across most metrics for both search and recommendation tasks at . Thus, we set for MT-Small dataset. After a similar experiment conducted on MT-Large dataset, we adopt the best-performing setting ().
(2) Loss weight of the intent translation contrastive learning (). To investigate the impact of our proposed intent translation contrastive learning, we vary in . Overall, the performance initially increases and then decreases with the increase of . Particularly, our model with set in {0.2, 0.4, 0.6} outperforms the version without translation contrastive learning on all metrics, demonstrating a proper loss weight of intent translation contrastive learning can aid in intent relation modeling. The optimal for search is 0.2 while for recommendation task, it is for Hit@5 and for NDCG@5. Therefore, we set for MT-Small dataset. Also, after conducting a similar experiment on MT-Large dataset, we adopt the best-performing setting .
6. Conclusion
This paper introduced a novel approach to unified intention-aware modeling for joint optimization of search and recommendation tasks. We recognized that user behaviors were motivated by their inherent intents and changing demand intents. To accurately learn users’ implicit demand intents for recommendation, we innovated a demand intent generator that utilized explicit queries from search data for supervised learning. Furthermore, we proposed a dual-intent translation propagation mechanism for interpretive modeling of the relation between users’ dual intents and their interactive items. In particular, we introduced an intent translation contrastive method to further constrain this relation. Our extensive offline experiments demonstrated that UDITSR outperformed the leading baselines in both search and recommendation tasks. Besides, online A/B tests further confirmed the superiority of our model. Finally, the intent visualization clearly explained the deeper reason for the remarkable improvement of our model.
Acknowledgements.
This research work is supported by the National Key Research and Development Program of China under Grant No.2021ZD0113602, the National Natural Science Foundation of China under Grant No.62176014 and No.62306255, the Fundamental Research Funds for the Central Universities and the Fundamental Research Project of Guangzhou under Grant No. 2024A04J4233.References
- (1)
- Ai et al. (2019) Qingyao Ai, Daniel N Hill, SVN Vishwanathan, and W Bruce Croft. 2019. A zero attention model for personalized product search. In Proceedings of the 28th ACM International Conference on Information and Knowledge Management. 379–388.
- Belkin and Croft (1992) Nicholas J Belkin and W Bruce Croft. 1992. Information filtering and information retrieval: Two sides of the same coin? Commun. ACM 35, 12 (1992), 29–38.
- Bi et al. (2020) Keping Bi, Qingyao Ai, and W Bruce Croft. 2020. A transformer-based embedding model for personalized product search. In Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval. 1521–1524.
- Bordes et al. (2013) Antoine Bordes, Nicolas Usunier, Alberto Garcia-Duran, Jason Weston, and Oksana Yakhnenko. 2013. Translating embeddings for modeling multi-relational data. Advances in neural information processing systems 26 (2013).
- Chen et al. (2019) Tong Chen, Hongzhi Yin, Hongxu Chen, Rui Yan, Quoc Viet Hung Nguyen, and Xue Li. 2019. Air: Attentional intention-aware recommender systems. In 2019 IEEE 35th International Conference on Data Engineering (ICDE). IEEE, 304–315.
- Cheng et al. (2016) Heng-Tze Cheng, Levent Koc, Jeremiah Harmsen, Tal Shaked, Tushar Chandra, Hrishi Aradhye, Glen Anderson, Greg Corrado, Wei Chai, Mustafa Ispir, et al. 2016. Wide & deep learning for recommender systems. In Proceedings of the 1st workshop on deep learning for recommender systems. 7–10.
- Covington et al. (2016) Paul Covington, Jay Adams, and Emre Sargin. 2016. Deep neural networks for youtube recommendations. In Proceedings of the 10th ACM conference on recommender systems. 191–198.
- Derr et al. (2018) Tyler Derr, Yao Ma, and Jiliang Tang. 2018. Signed graph convolutional networks. In 2018 IEEE International Conference on Data Mining (ICDM). IEEE, 929–934.
- Donahue et al. (2014) Jeff Donahue, Yangqing Jia, Oriol Vinyals, Judy Hoffman, Ning Zhang, Eric Tzeng, and Trevor Darrell. 2014. Decaf: A deep convolutional activation feature for generic visual recognition. In International conference on machine learning. PMLR, 647–655.
- Fan et al. (2022) Lu Fan, Qimai Li, Bo Liu, Xiao-Ming Wu, Xiaotong Zhang, Fuyu Lv, Guli Lin, Sen Li, Taiwei Jin, and Keping Yang. 2022. Modeling user behavior with graph convolution for personalized product search. In Proceedings of the ACM Web Conference 2022. 203–212.
- Feng et al. (2019) Yufei Feng, Fuyu Lv, Weichen Shen, Menghan Wang, Fei Sun, Yu Zhu, and Keping Yang. 2019. Deep session interest network for click-through rate prediction. In Proceedings of the 28th International Joint Conference on Artificial Intelligence. 2301–2307.
- Ferri et al. (2011) Cesar Ferri, José Hernández-Orallo, and Peter A Flach. 2011. A coherent interpretation of AUC as a measure of aggregated classification performance. In Proceedings of the 28th International Conference on Machine Learning (ICML-11). 657–664.
- Guo et al. (2017) Huifeng Guo, Ruiming Tang, Yunming Ye, Zhenguo Li, and Xiuqiang He. 2017. DeepFM: a factorization-machine based neural network for CTR prediction. In Proceedings of the 26th International Joint Conference on Artificial Intelligence. 1725–1731.
- He et al. (2020) Xiangnan He, Kuan Deng, Xiang Wang, Yan Li, Yongdong Zhang, and Meng Wang. 2020. Lightgcn: Simplifying and powering graph convolution network for recommendation. In Proceedings of the 43rd International ACM SIGIR conference on research and development in Information Retrieval. 639–648.
- He et al. (2017) Xiangnan He, Lizi Liao, Hanwang Zhang, Liqiang Nie, Xia Hu, and Tat-Seng Chua. 2017. Neural collaborative filtering. In Proceedings of the 26th international conference on world wide web. 173–182.
- Hu et al. (2008) Yifan Hu, Yehuda Koren, and Chris Volinsky. 2008. Collaborative filtering for implicit feedback datasets. In 2008 Eighth IEEE international conference on data mining. Ieee, 263–272.
- Järvelin and Kekäläinen (2002) Kalervo Järvelin and Jaana Kekäläinen. 2002. Cumulated gain-based evaluation of IR techniques. ACM Transactions on Information Systems (TOIS) 20, 4 (2002), 422–446.
- Kipf and Welling (2016) Thomas N Kipf and Max Welling. 2016. Semi-Supervised Classification with Graph Convolutional Networks. In International Conference on Learning Representations.
- Li et al. (2010) Lihong Li, Wei Chu, John Langford, and Robert E Schapire. 2010. A contextual-bandit approach to personalized news article recommendation. In Proceedings of the 19th international conference on World wide web. 661–670.
- Li et al. (2021) Sen Li, Fuyu Lv, Taiwei Jin, Guli Lin, Keping Yang, Xiaoyi Zeng, Xiao-Ming Wu, and Qianli Ma. 2021. Embedding-based product retrieval in taobao search. In Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining. 3181–3189.
- Lian et al. (2018) Jianxun Lian, Xiaohuan Zhou, Fuzheng Zhang, Zhongxia Chen, Xing Xie, and Guangzhong Sun. 2018. xdeepfm: Combining explicit and implicit feature interactions for recommender systems. In Proceedings of the 24th ACM SIGKDD international conference on knowledge discovery & data mining. 1754–1763.
- Liu et al. (2020a) Shang Liu, Wanli Gu, Gao Cong, and Fuzheng Zhang. 2020a. Structural relationship representation learning with graph embedding for personalized product search. In Proceedings of the 29th ACM International Conference on Information & Knowledge Management. 915–924.
- Liu et al. (2020b) Zhiwei Liu, Xiaohan Li, Ziwei Fan, Stephen Guo, Kannan Achan, and S Yu Philip. 2020b. Basket recommendation with multi-intent translation graph neural network. In 2020 IEEE International Conference on Big Data (Big Data). IEEE, 728–737.
- Mao et al. (2021) Kelong Mao, Jieming Zhu, Jinpeng Wang, Quanyu Dai, Zhenhua Dong, Xi Xiao, and Xiuqiang He. 2021. SimpleX: A simple and strong baseline for collaborative filtering. In Proceedings of the 30th ACM International Conference on Information & Knowledge Management. 1243–1252.
- Niu et al. (2020) Xichuan Niu, Bofang Li, Chenliang Li, Rong Xiao, Haochuan Sun, Hongbo Deng, and Zhenzhong Chen. 2020. A dual heterogeneous graph attention network to improve long-tail performance for shop search in e-commerce. In Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 3405–3415.
- Pi et al. (2020) Qi Pi, Guorui Zhou, Yujing Zhang, Zhe Wang, Lejian Ren, Ying Fan, Xiaoqiang Zhu, and Kun Gai. 2020. Search-based user interest modeling with lifelong sequential behavior data for click-through rate prediction. In Proceedings of the 29th ACM International Conference on Information & Knowledge Management. 2685–2692.
- Qin et al. (2023) Chuan Qin, Le Zhang, Rui Zha, Dazhong Shen, Qi Zhang, Ying Sun, Chen Zhu, Hengshu Zhu, and Hui Xiong. 2023. A comprehensive survey of artificial intelligence techniques for talent analytics. arXiv preprint arXiv:2307.03195 (2023).
- Radev et al. (2002) Dragomir R Radev, Hong Qi, Harris Wu, and Weiguo Fan. 2002. Evaluating Web-based Question Answering Systems.. In LREC. Citeseer.
- Ren et al. (2023) Xubin Ren, Lianghao Xia, Jiashu Zhao, Dawei Yin, and Chao Huang. 2023. Disentangled Contrastive Collaborative Filtering. In Proceedings of the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval (, Taipei, Taiwan,) (SIGIR ’23). Association for Computing Machinery, New York, NY, USA, 1137–1146. https://doi.org/10.1145/3539618.3591665
- Rendle et al. (2009) Steffen Rendle, Christoph Freudenthaler, Zeno Gantner, and Lars Schmidt-Thieme. 2009. BPR: Bayesian personalized ranking from implicit feedback. In Proceedings of the Twenty-Fifth Conference on Uncertainty in Artificial Intelligence. 452–461.
- Sarwar et al. (2001) Badrul Sarwar, George Karypis, Joseph Konstan, and John Riedl. 2001. Item-based collaborative filtering recommendation algorithms. In Proceedings of the 10th international conference on World Wide Web. 285–295.
- Scarselli et al. (2008) Franco Scarselli, Marco Gori, Ah Chung Tsoi, Markus Hagenbuchner, and Gabriele Monfardini. 2008. The graph neural network model. IEEE transactions on neural networks 20, 1 (2008), 61–80.
- Sondhi et al. (2018) Parikshit Sondhi, Mohit Sharma, Pranam Kolari, and ChengXiang Zhai. 2018. A taxonomy of queries for e-commerce search. In The 41st International ACM SIGIR Conference on Research & Development in Information Retrieval. 1245–1248.
- Sun et al. (2019) Fei Sun, Jun Liu, Jian Wu, Changhua Pei, Xiao Lin, Wenwu Ou, and Peng Jiang. 2019. BERT4Rec: Sequential recommendation with bidirectional encoder representations from transformer. In Proceedings of the 28th ACM international conference on information and knowledge management. 1441–1450.
- Tai et al. (2021) Chang-You Tai, Liang-Ying Huang, Chien-Kun Huang, and Lun-Wei Ku. 2021. User-centric path reasoning towards explainable recommendation. In Proceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval. 879–889.
- Veličković et al. (2018) Petar Veličković, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Liò, and Yoshua Bengio. 2018. Graph Attention Networks. In International Conference on Learning Representations.
- Wang et al. (2019b) Shoujin Wang, Liang Hu, Yan Wang, Quan Z Sheng, Mehmet Orgun, and Longbing Cao. 2019b. Modeling multi-purpose sessions for next-item recommendations via mixture-channel purpose routing networks. In International Joint Conference on Artificial Intelligence. International Joint Conferences on Artificial Intelligence.
- Wang et al. (2020) Shoujin Wang, Liang Hu, Yan Wang, Quan Z Sheng, Mehmet Orgun, and Longbing Cao. 2020. Intention2basket: A neural intention-driven approach for dynamic next-basket planning. In Twenty-Ninth International Joint Conference on Artificial Intelligence and Seventeenth Pacific Rim International Conference on Artificial Intelligence IJCAI-PRICAI-20. International Joint Conferences on Artificial Intelligence Organization.
- Wang et al. (2019a) Xiang Wang, Xiangnan He, Meng Wang, Fuli Feng, and Tat-Seng Chua. 2019a. Neural graph collaborative filtering. In Proceedings of the 42nd international ACM SIGIR conference on Research and development in Information Retrieval. 165–174.
- Wu et al. (2022) Shiwen Wu, Fei Sun, Wentao Zhang, Xu Xie, and Bin Cui. 2022. Graph neural networks in recommender systems: a survey. Comput. Surveys 55, 5 (2022), 1–37.
- Wu et al. (2020) Zonghan Wu, Shirui Pan, Fengwen Chen, Guodong Long, Chengqi Zhang, and S Yu Philip. 2020. A comprehensive survey on graph neural networks. IEEE transactions on neural networks and learning systems 32, 1 (2020), 4–24.
- Yan et al. (2018) Sijie Yan, Yuanjun Xiong, and Dahua Lin. 2018. Spatial temporal graph convolutional networks for skeleton-based action recognition. In Proceedings of the AAAI conference on artificial intelligence, Vol. 32.
- Yao et al. (2021) Jing Yao, Zhicheng Dou, Ruobing Xie, Yanxiong Lu, Zhiping Wang, and Ji-Rong Wen. 2021. USER: A unified information search and recommendation model based on integrated behavior sequence. In Proceedings of the 30th ACM International Conference on Information & Knowledge Management. 2373–2382.
- Zamani and Croft (2018) Hamed Zamani and W Bruce Croft. 2018. Joint modeling and optimization of search and recommendation. arXiv preprint arXiv:1807.05631 (2018).
- Zamani and Croft (2020) Hamed Zamani and W Bruce Croft. 2020. Learning a joint search and recommendation model from user-item interactions. In Proceedings of the 13th international conference on web search and data mining. 717–725.
- Zhang et al. (2019) Chuxu Zhang, Dongjin Song, Chao Huang, Ananthram Swami, and Nitesh V Chawla. 2019. Heterogeneous graph neural network. In Proceedings of the 25th ACM SIGKDD international conference on knowledge discovery & data mining. 793–803.
- Zhao et al. (2022) Kai Zhao, Yukun Zheng, Tao Zhuang, Xiang Li, and Xiaoyi Zeng. 2022. Joint learning of e-commerce search and recommendation with a unified graph neural network. In Proceedings of the Fifteenth ACM International Conference on Web Search and Data Mining. 1461–1469.
- Zhou et al. (2019) Guorui Zhou, Na Mou, Ying Fan, Qi Pi, Weijie Bian, Chang Zhou, Xiaoqiang Zhu, and Kun Gai. 2019. Deep interest evolution network for click-through rate prediction. In Proceedings of the AAAI conference on artificial intelligence, Vol. 33. 5941–5948.
- Zhou et al. (2018) Guorui Zhou, Xiaoqiang Zhu, Chenru Song, Ying Fan, Han Zhu, Xiao Ma, Yanghui Yan, Junqi Jin, Han Li, and Kun Gai. 2018. Deep interest network for click-through rate prediction. In Proceedings of the 24th ACM SIGKDD international conference on knowledge discovery & data mining. 1059–1068.
- Zhu et al. (2020) Nengjun Zhu, Jian Cao, Yanchi Liu, Yang Yang, Haochao Ying, and Hui Xiong. 2020. Sequential modeling of hierarchical user intention and preference for next-item recommendation. In Proceedings of the 13th international conference on web search and data mining. 807–815.