Papers^γ

# Summary ### Introduction ##### about hybrid reco : - many RS have an underlying HIN structure and are achieving hybrid reco (in the sense using both user feedbacks and content-based info) ##### difference with existing methods : - use different types of relations, benefit : use the fact that users consume an item for different reasons (e.g. : movie for genre, for director, etc) ##### Recommender System : - combine users feedback and various types of info in a collaborative filtering style - use metapath in HIN to generate reco - technical implementation uses Matrix Factorization ##### Datasets : - MovieLens 100K combined with IMDB and Yelp, implicit feedback only ##### Contributions: - study reco with implicit feedback in HIN - use network heterogeneity to spread preferences on the metapaths - generate personalized reco - specific case study : ML100K and Yelp ### Background and preliminaries ##### binary user feedback - explain how to generate the bipartite adjacency matrix ##### Heterogeneous Information Network - definition (using entity mapping function and link entity mapping function) - vocabulary to describe HIN ##### Matrix Factorization for implicit feedback - describe principle of MF (decomposing the feedback matrix) - resolution using NMF ##### Problem definition - how to make personalized recommendation based on implicit feedback in the form of a list of recommendations ### Meta-path based latent features ##### meta-path - definition and interest (types of paths in a HIN) - can be used to measure similarity and proximity between entities - ex: user [watches] movie [watched by] user [following] actor [starring] movie ##### user preference diffusion - type of meta-paths considered in the paper : user -> item -> * -> item (* may be tag, genre, director, plot for ML100K ; * may be category, customer, location for Yelp) - define user preference score : normalized weighted sum of the number of paths to a given item (eq 2) - if L types of metapaths, then L matrices R (user preference matrices) - use these scores to build the recommendation model ##### global recommendation model - define the recommendation mechanic which is inspired by MF - (?) MF may be achieved on each user preference matrix taken separately : find a couple of reduced matrices with NMF, then prediction model is given by equation 4 - RK: not personalized as coefficients are the same for every user ### Personalized recommendation model - same principle as global recommendation method, except that there is first a clustering, and the learning is achieved cluster per cluster - the number of clusters is a parameter of the method ### Model learning with implicit feedback - learning the model parameters (thetas in equation 4) - use implicit feedback to do so (1 = user browses item / 0 = user does not) - usually prediction done with either classification or learning-to-rank but their approach: rank 1s above 0s (in the spirit of ref 21) ##### Bayesian ranking-based optimization - assumption: a user ranking is independent from the others (allow to get eq.7) - assumption on the probability expressed in equation 8 - allows to derive the expression of objective function O ##### optimization algorithm - optimization: finds thetas such that dO/dTheta = 0 - method: Stochastic Gradient Descent ##### learning personalized recommendation models - this technique is not personalized - to personalize the reco: clusters with a k-means method ### Empirical study ##### Data - dataset 1 : IMDB + ML100K (IM100K); if user has seen movie 1 else 0 - dataset 2 : Yelp; if user has reviewed buisness 1 else 0 - d2 much sparser than d1 (see feedback distribs on figure 5) - temporal split 80% / 20% between training and test ##### Competitors and evaluation metrics - RS benchmarks: popularity-based, co-click, NMF (baseline of collaborative filtering), hybrid SVM - for their method: 10 different metapaths différentes (see Table 6.2) - evaluation: as based on implicit feedback, precision at position and top-10 mean reciprocal rank (MRR_k) ##### Performance comparison - Table 3 for a summary - very few items interact with a lot of users - parameters for NMF: dimension of the reduced matrix: 20 (IM100K), 60 (Yelp) - Hybrid-SVM uses the same info as their method (HeteRec) and uses PathSim - in general HeteRec better than all benchmark methods - in particular HeteRec > Hybrid-SVM (while similar information) - improvement higher for Yelp than for IM100K, possibly a consequence of Yelp sparsity - HeteRec-p (personalized version) : even better than HeteRec-g ##### Performance analysis - more precise analysis of the performances on IM100K only for HeteRec-g , HeteRec-p , NMF , Co-Click - divide in 6 different training sets, depending on various parameters - performances increase with the number of movies watched for all methods except co-click - performances decrease with movies popularity for all methods except co-click ##### Parameter tuning - HeteRec have more parameters - regularization parameter lambda (eq 9) computed with cross-validation - sampling necessary for Yelp (as 10^12 elements), performance variations with sampling represented on Fig7 (relatively stable) - for HeteRec-p: number of clusters, see fgure 6c ### Related work ##### CF based hybrid RS ##### Information network analysis

Very good article from GroupLens team about the diversity narrowing effect and the role of the recommender system. Experimental measurements on the MovieLens platform, the content of movies is described with tag-genome, the RS is an item-item collaborative filtering. ### Introduction ##### two research questions : - Do recommender systems expose users to narrower content over time? - How does the experience of users who take recommendations differ from that of users who do not regularly take recommendations? ##### method - users in two categories (following / ignoring ~ control group) - then sort of A/B testing to measure consumption diversity and enjoyment at the individual level ##### 4 contributions : - method to analyze the effects of RS on users - give quantitative proofs that users using RS have a better experience than others - show that diversity reduction effect is small - show that users using RS tend to consume more divers contents ### Related work - Pariser (11) on filter bubble - Tetlock (16) measures bias induce by bubbles (among economists) - Sunstein (15) thinks that personnalization tends to reduce the space of shared experience among users - Negroponte (MIT MediaLab co-founder) defends that algorithms may also open horizons - Linden (Amazon RS contributer, 5) thinks that reco can generate some form of serendipity and oppose to Pariser's thesis - Fleder et al (2) use simulation to show that RS tend to uniform experience / Hosanagar et al (3) use measurements, but limitations to these studies, e.g. simplistic model of human behavior (Fleder) ### Data & metrics ##### Dataset MovieLens (September 2013) : - 220.000 users - 20M ratings, 20000 movies - RS (item-item Collaborating Filtering, similar to Ama) propose "top picks" per user (15 default value) ##### Tag-genome to describe movie content - information space which describes movies with tags given by users - 9500 movies in tag-genome (april 2013) and 1100 tags => 10-11M pairs ##### Time period - 21 months from Feb 2008 to Aug 2010 (because less missing data) ##### preprocessing : - 15 first ratings taken out (platform propositions) - then first 3 months taken out (as many ratings are "catching up" + to give users time to get used to ML and ML to get info from the user ##### recommendation blocks definition : - several possibilities: per login session, per periods of time, but as users don't have the same activity - => blocks of 10 consecutive ratings, 10 is roughly the median of number of ratings per 3 months ##### users and items : - only users who have their first rating during the period of interest and who have at least 3 rating blocks => 1400 users with 3 to 200 rating blocks - 173,000 ratings on 10500 movies, all in tag genome (small inconsistency: there is only 9500 movies in the tag-genome ##### Identifying consumed recommendations in a rating block - groups based on the number of reco followed by users - criterion to consider that reco followed: in the list of top picks between 3h and 3 months before evaluation ##### Ignoring group vs following group - 2 groups of users - pb: some users take a lot of reco during a period and then very few - => rank users depending on the proportion of rating blocks during which they followed a reco (Fig4) - then if >50% : following group (286 users); if 0% : ignoring group (430 users) (seems quite ad hoc, but makes sense as differences are not very large and following reco must be quite rare) ##### Measuring content diversity: tag genome (see figure 5) - matrix movie-tag with 1 to 5 relevance score - movie is then a vector of scores - distance between movies is euclidean distance in this space (rather than cosine similarity because of matrix density) - order of magnitudes: min distance 5.1 (Halloween 4 - Halloween 5), max distance 44.2 (Matrix - Paris was a woman), mean 23.4 - authors defend that tag-genome is very expressive (more than cosine similarity), also benefits from the continuous input from users (illustrate on examples) ##### Measuring the effect of recommender systems ###### # content diversity metrics (standard) - average pairwise distance of the movies in the list - maximum pairwise distance of the movies in the list - on recommended movies (15 top picks) - on rated movies - more or less normally distributed (see fig6) ###### # user experience : - average rating per user - more or less normally distributed ###### # analysis : - mean shift for users in both groups - measure difference between groups and within group at different times - within group : use standard t-test (as same size) - between groups : use Welch t-tests (as different sizes) ### Results ##### Research Question 1: do RS expose users to narrower content over time ? - diversity on recommended movies: see tab2 - statistically significant drop on following group - significant drop (?) on ignoring group - following more diverse than ignoring (significant), but gap narrows over time ##### Research Question 2: does the experience of users who take recommendations differ from that of users who do not regularly take recommendations ? - diversity on rated movies using mean distance: see tab3 - beginning: no significant difference - end: significant drop for both groups - diversity on rated movies using max distance: see tab4 - same trend - enjoyment evaluated with rating between first and last block for both groups: see tab5 - following group gives higher ratings, and the rating drop is lower in following than in ignoring group - similar trend with rating means (see tab7) - refined analysis in tab6 with ratings depending on the fact that movies rated are recommended or not => better experience if movie recommended (complement) - rank together all blocks of all groups at all times per increasing average rating - => block located in this ranking with its percentile (ex: first block, ignoring group ~ 63rd percentile) - in tab8: percentile drop for both groups => percentile drop for ignoring is high (-19) not in the following (-1) ### Discussion ##### summary - diversity tends to narrow in anyway over time - effect subdued for the group which follows reco - users following reco get more diverse recommendations - ratings seem to encourage RS to broaden recommendations ##### prospects - is there a natural trend to narrow consumption diversity? - they think that item-item CF slows this effect down - the RS can also inform the user of his or her consumption diversity - finally RS can be designed to intentionally force diversity ##### limitations - restriction to top picks - restriction to one dataset and one system (item-item CF) (note: on current platform (May 2019), 4 different RS: peasant/bard/warrior/wizard, the one described in the article seems to be warrior, default RS)

Very interesting article. I'd like to get their data on the "Top Picks for you". There's probably a correlation between movie releases and what users watch and what pop into Top Picks for you, source of (minor?) bias. They use one metric for content diversity, there's room to explore other ones. They use Ziegler's work on intra list diversity ([21]).

Continuing the series of articles (... well actually is just the second one) on restricted combinatorial objects constrained by a recursively defined statistic and initiated by [this work](https://papers-gamma.link/paper/38/Dyck%20paths%20with%20a%20first%20return%20decomposition%20constrained%20by%20height) we submitted to review a new paper of subject.

Il y a Lukasuewicz à faire maintenant ... 😀