Learning Global Term Weights for Content-based Recommender Systems

Abstract

Recommender systems typically leverage two types of signals to effectively recommend items to users: user activities and content matching between user and item profiles, and recommendation models in literature are usually categorized into collaborative filtering models, content-based models and hybrid models. In practice, when rich profiles about users and items are available, and user activities are sparse (cold-start), effective content matching signals become much more important in the relevance of the recommendation. The de-facto method to measure similarity between two pieces of text is computing the cosine similarity of the two bags of words, and each word is weighted by TF (term frequency within the document) × IDF (inverted document frequency of the word within the corpus). In general sense, TF can represent any local weighting scheme of the word within each document, and IDF can represent any global weighting scheme of the word across the corpus. In this paper, we focus on the latter, i.e., optimizing the global term weights, for a particular recommendation domain by leveraging supervised approaches. The intuition is that some frequent words (lower IDF, e.g. "database") can be essential and predictive for relevant recommendation, while some rare words (higher IDF, e.g. the name of a small company) could have less predictive power. Given plenty of observed activities between users and items as training data, we should be able to learn better domain-specific global term weights, which can further improve the relevance of recommendation. We propose a unified method that can simultaneously learn the weights of multiple content matching signals, as well as global term weights for specific recommendation tasks. Our method is efficient to handle large-scale training data * This work was conducted during an internship at LinkedIn. Copyright is held by the International World Wide Web Conference Committee (IW3C2). IW3C2 reserves the right to provide a hyperlink to the author's site if the Material is used in electronic media.

HealthRecSys: A semantic content-based recommender system to complement health videos

“…Recommender systems reflect the user’s interest and make proper personalized recommendation through several methods. Most current systems have adopted recently developed algorithms that use machine-learning [ 14 – 16 ], naive Bayes [ 16 , 17 ], social-trust-based [ 18 – 21 ], constraint-based [ 22 ], case-based [ 23 , 24 ], and matrix factorization [ 25 , 26 ] approaches. Recommender systems are also found in clinical settings, mainly to assist health professionals, though some systems assist family members, patients, or caregivers [ 27 – 29 ].…”

A Novel Recommendation System to Match College Events and Groups to Students

“…• Collaborative recommendation systems: These approaches recommend the items based on the similarity of tastes and preferences between users using a utility matrix [4,5]. • Content-based recommendation systems: These approaches recommend the items based on the description of the item and a profile of the user's preference and history [6][7][8][9][10]. • Hybrid recommendation systems: These approaches combine collaborative and content-based methods to recommend some items to the users [11,12].…”

“…Gu et al [7] proposed a method to measure the similarity between two pieces of text using cosine similarity of the two bags of words, where each word is weighted by its tf-idf, and they applied this method on LinkedIn job data. A similar method has been proposed by Philip et al [8] but for digital libraries. Also, Lak et al [9] proposed an article recommendation system that uses tf-idf and word2vec for converting each article to a feature-vector.…”

Content-Based Top-N Recommendation Using Heterogeneous Relations

“…A standard way to measure similarity between two profiles is computing the cosine similarity of the two bags of words, and each word is weighted by term frequency tf (within the document) × inverted document frequency idf (of the term within the corpus). While the local term frequency could be computed offline, it has been suggested [9] that the global term weights idf requires further optimization to achieve better precision. Thus, we optimize the global term weights with the guidance of the similarity from PathSim.…”