Exploiting Cross-Channel Information for Personalized Recommendation
COM2 Level 2
MR3, COM2-02-26
Abstract:
In the era of information overload, recommender system has gained widespread adoption across industry to drive various online customer-oriented services. It facilitates users to discover a small set of relevant items, which meet their personalized interests, from overwhelming choices. Generally, the modeling of user-item interactions is at the heart of personalized recommendation. Nowadays, diverse kinds of auxiliary information on users and items become increasingly available in online platforms, such as user demographics, social relations, and item knowledge. More recent evidence suggests that incorporating such auxiliary data with collaborative filtering can better capture the underlying and complex user-item relationships, and further achieve higher recommendation quality. Furthermore, such auxiliary data, such as social relations and deep knowledge on items, enable us to uncover valuable evidence as well as reasons on why a recommendation is made.
In this thesis proposal, we investigate techniques to exploit auxiliary information, especially those across different channels, for enriching the representations of users and items, as well as modeling the user-item interactions more comprehensively. As defined in cross-channel marketing, the channel refers to the platform that a business can use to reach potential customers, including online platforms and offline retail stores. Different channels emphasize different aspects or behaviors of users. From the viewpoint of recommendation, we define the channel as one domain or source which can indicate users' preferences towards an item. Consequently, cross-channel means incorporating all the different sources and establishing the whole view of user-item interactions. Therefore, we can classify rich auxiliary information into various types of channels in different recommendation scenarios.
First, We have specifically focused on leveraging user behaviors across online and offline channels for event recommendation, utilizing connections between information- and social-oriented channels for cross-domain recommendation, and exploring cross features incorporating the user-centric and item-centric channels. Particularly, for event recommendation, we unified users' online and offline behaviors into user representations, to better depict user preferences on target events.
Second, we investigated a novel task of cross-channel social recommendation, which aims to recommend relevant items of information channel to potential users within the social channels. While integrating such data sources is able to offer strong predictive performance, important questions are being raised on why the recommended items are suitable for the user. Towards this end, we worked on the explainable recommendation task, where top cross features ranked by attention weights can be treated as concrete reasons for a recommendation. Here cross features incorporate the basic features derived from user-centric and item-centric channels, representing the general decision rules.
Third, we have incorporated knowledge-aware channels, especially knowledge graph, into recommender systems, aiming to inject semantics into the representation learning of user and item representations. Furthermore, the usage of such knowledge endows recommender systems strong representation ability and reasoning power. As such, we are capable of making more accurate recommendations results with knowledge-aware explanations, exhibiting the process of information propagation.