Explainable Cross-Domain Recommendations Through Relational Learning

We propose a method to generate explainable recommendation rules on cross-domain problems. Our two main contributions are: i) using relational learning to generate the rules which are able to explain clearly why the items were recommended to the particular user, ii) using the user’s preferences of items on different domains and item attributes to generate novel or unexpected recommendations for the user. To illustrate that our method is indeed feasible and applicable, we conducted experiments on music and movie domains. Recommender systems (RS) currently become one of the most basic supportive techniques in an online landscape/world. RS has proven to be a major source of enhanced functionality, user satisfaction, and revenue improvement. The most common critical issues found with RS include maximizing prediction accuracy, solving cold-start problem, reducing sparsity, providing novelty, diversity and serendipity. However, solving one problem may create another problem, or a trade-off. All issues have not been perfectly solved since many current recommender algorithms seem to be locked away inside a black box. Once an algorithm is processed, it is quite difficult to understand why it gives a particular recommendation to a set of data inputs. If we can understand the reason behind the recommendation, we believe we will be able to possibly find the way to handle such problems more effectively. Cross-domain approach improves prediction accuracy by reducing data sparsity and offering added values to recommendations by providing diversity, novelty and serendipity predictions (Cantador et al. 2015). In cross-domain recommendation tasks, the systems recommend items in the target domain to users in the source domain. There are two types of cross-domain approach: Aggregating and Transferring and they are different mainly based on how knowledge from the source domain is exploited. Relational learning has already shown its use in RS. The evidences from the researches by Kouki et al. (2015) and Catherine and Cohen (2016) indicate that relational learning provides a better recommendation performance by incorporating additional information compared to traditional methods with a single dyadic relationship between the objects, i.e. Copyright c © 2018, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved. users and items. Hence, the relational learning captured our interest to model and provide a potential solution for explainable cross-domain recommendations. We propose a method to generate explainable recommendation rules on cross-domain problems using relational learning. The generated rules explain why the system gives a particular recommendation to a user. The rules are simple and understandable. Moreover, some novel recommendations in the primary domain are generated based on the user’s preference on additional domains. Research Methodology We first investigated how a user’s taste of another type-level item (e.g., movie, book, game) is used to predict the user’s music taste, for example, could a user’s movie taste be used to predict his music taste? and how? Our initial experiments were mainly conducted over two datasets, user’s music and movie preferences and item attributes, obtained from Amazon product dataset provided by UCSD.1 The user preference dataset included reviews, product metadata and links. The item attribute dataset contained attributes of music and movies listed in the first dataset. This dataset was developed using Amazon Product API. Our two proposed methods used to generate the music recommendation rules are described below. Method1: Recommendations on cross domains using items. This method was designed based on traditional recommendation approach. Pearson correlation was used to find similarities between users, then deduce the certain recommendation rules. For the user-based recommendation, the similar taste on movies would lead to a similar taste on music. For the itembased recommendation, the high confidence of the defined rule would result in a music recommendation. The ProbLog2 syntax was used to describe the model. The approach was performed in three steps: define rule, query and select output. Only output greater than or equal to the desired threshold would be selected, some examples are shown below. Example: User-based recommendation %"Music" will be recommended to user2 if user1 likes %that "Music" and his movie preference influences user2 Defined rule: likedMusic(U2,Music) :influencesUser(U1,U2), likedMusic(U1,Music). Query: query(likedMusic(U2,Music)). http://jmcauley.ucsd.edu/data/amazon/ https://dtai.cs.kuleuven.be/problog/ The Thirty-Second AAAI Conference on Artificial Intelligence (AAAI-18)