Language Models of Collaborative Filtering

Abstract

Abstract. Collaborative filtering is a major technique to make personalized recommendations about information items (movies, books, webpages etc) to individual users. In the literature, a common research objective is to predict unknown ratings of items for a user, on the condition that the user has explicitly rated a certain amount of items. Nevertheless, in many practical situations, we may only have implicit evidence of user preferences, such as "playback times of a music file" or "visiting frequency of a web-site". Most importantly, a more practical view of the recommendation task is to directly generate a top-N ranked list of items that the user is most likely to like. In this paper, we take these two concerns into account. Item ranking in recommender systems is considered as a task highly related to document ranking in text retrieval. Firstly, two practical item scoring functions are derived by adopting the generative language modelling approach of text retrieval. Secondly, to address the uncertainty associated with the score estimation, we introduce a risk-averse model that penalizes the less reliable scores. Our experiments on real data sets demonstrate that significant performance gains have been achieved.

Relevance-based language modelling for recommender systems

“…However, when modelling the recommendation problem, Dirichlet can suffer from the undesired effect of demoting those items that have been recently introduced in the system and so have very few recommendations. In fact, in (Wang, 2009) the smoothing for the LM based recommendation with Dirichlet smoothing presents significantly worse performance than using Jelinek-Mercer in one of the experiments reported there. For estimating p(i) we decided to keep it simple and a uniform distribution was chosen.…”

“…In (Wang et al, 2008b), the authors found interesting analogies between CF with implicit data (where the evidence of user interest for items consists of access frequencies, rather than explicit preference rating values) and IR, introducing the concept of binary relevance into CF and applying the Probability Ranking Principle of IR to CF. Similarly, in (Wang et al, 2006a) a generative relevance model is proposed for implicit CF, and in (Wang, 2009), the author made use of a language modelling formulation to propose a risk-aware ranking for implicit CF. The approach we propose here is much in tune with the spirit of this line of research on model unification.…”

Exploring Statistical Language Models for Recommender Systems

“…An advantage of the Information Retrieval methods is that they are traditionally focused on generating a ranked list of items. Thus, there exists an emerging interest in applying IR techniques to the field of Recommender Systems [18,17,5,13].…”

“…Another approach is the proposal of Wang et al based on the probability ranking principle [18]. Wang also derived a CF method utilising Language Models using a risk-averse model that penalises less reliable scores [17]. Nevertheless, these methods are intended to employ implicit feedback.…”

A Study of Smoothing Methods for Relevance-Based Language Modelling of Recommender Systems

“…the state of the art methods. Following previous results in using Language Models for the recommendation task [6], the authors decided to use Jelinek-Mercer for smoothing the different probabilities arguing that Dirichlet priors can demote the weight of those items recently introduced in the systems. In this paper we tested those intuitions and analysed the performance of different smoothing techniques in the recommendation task.…”