Papers^γ

### Short summary ##### Plan - procedure for article selection - recommender systems overview - review itself: diversity measure ; impact of diversification on recommendation ; diversification methods - conclusion and perspectives ##### Procedure for article selection - search on Google Scholar with keywords selection - doubles elimination - selection of articles without additional payments - clustering into three groups of articles (based on review plan) ##### RS overview (standard, aiming at people new to the field) - dates back to Salton and McGill, 1980 (ref 1) - usual standard techniques: word vectors, DT, Naïve Bayes, kNN, SVM - applications: digital TV, web multimedia (YouTube, Shelfari (now merged into Goodreads), Facebook, Goodreads), personalized ads, online shopping - the general process of recommendation: past users activity collection ; create user model ; present recommendation information ; feedback collection (distinguishing explicit and implicit recommendation) - important challenges: data sparsity (working with "mostly empty user-items datasets") ; cold start (new users or items in the dataset ; overfitting (actually, rather in the sense of overspecialization) ##### Diversification - Table 1 summarizes diversity measures - Bradley-Smyth 2001: average dissimilarity between all pairs of items - Fleder-Hosanagar 2007: Gini - explore with a model how diversity evolves through recommendation cycles - Clarke et al. 2008: combined measure (ambiguity, redundancy, novelty...) - Vargas et al. 2011: intralist diversity - Hu-Pu 2011: perceived diversity (questionnaire) - Vargas et al. 2012: in the line of Clarke et al. 2008 - Castagnos et al. 2013: in the line of Bradley-Smyth 2001 - develop experiments with users - L'Huillier et al. 2014: idem - Vargas et al. 2014: binomial diversity (mixing coverage and redundancy) - Table 2 summarizes how diversity affects recommendation - usually: F-measure, MAE, NMAE - some articles show that diversification by reranking is possible without affecting too much accuracy (ex: Adomavicius et Kwon, 51) - some address the question of trade-off between diversity and accuracy (52: Hurley-Zhang 2011, 55: Aytekin-Karakaya 2014, 56: Ekstrand et al, 2014, 57: Javari-Jalili, 2014) - pb seen as multi-objective, looking for Pareto efficient ranking (58: Ribeiro et al 2015) - Table 3 summarizes diversification algorithms - many methods are reranking from accuracy-based ranking (59: Ziegler et al. 2005, 51 et 61: Adomavicius-Kwon 2012, 2011, 62: Premchaiswadi et al 2013) - then various strategies, depending on the method, on the type of data, whether the authors question temporal aspects - underlying idea is that the original algorithm (typically CF) is already diverse, just needs reordering ##### Conclusions - no consensus on a diversity metric - increasing diversity does not necessarily means sacrifice accuracy - various challenges: not enough live studies ; work in psychology would be useful ; how to use systems which have a lot of different types of items ; how to diversify during the reco process (and not a posteriori)

# In short CIKM 2015 article, one of the core reference on HIN-based recommendation, they consider a weighted variation which allows them to deal with rated cases. # Summary ### 1. Introduction ##### context and problem - more and more research on HIN for data mining: similarity search, clustering, classification - recently on reco: 2 (Jamali et al), 7 (Luo et al), 15 (Sun et al): interesting because possibility to integrate various kind of information (cf Fig 1), ex: user-movie-user is more or less a CF model - pb1: no values on links while recommendations are often associated with ratings (ex: Tom and Mary on Fig1, who have seen different movies but very different ratings) => necessary to generalize the formalism of HIN in order to account for the link weights - pb2: problem to combine information from different meta-paths, a good method of weight learning should allow to obtain personalized weights, weights should contribute to explanation, but if we personalize recommendation data sparsity problems get worst - contributions: extend HINs to weighted case ; propose SemRec (semantic path based RS) which flexibly include information of heterogeneous nature ; define consistency rule of similar users to circumvent data sparsity pb ; empirically study 2 datasets: Yelp and Douban ### 2. Heterogeneous network framework for recommendation ##### 2.1 Basic concepts - HIN for the weighted case (as usual, but with weights on one or more relation) ; illustration from Fig 2a - extended meta-paths to paths with attributes: links weights must be in a given range (give illustration) ##### 2.2 Recommendation on HIN - Table1: semantic example associated to meta-paths - Discussion about how different RS models will use meta-paths ##### 2.3 Similarity measure based on weighted meta-path - go through literature reco models based on paths in HIN (but no WHIN): PathSim (12), PCRW (4), HeteSim(10) ; cannot be simply transposed as they have no notion of weight - they adapt to this context with "atomic meta path": meta-path where weights take specific values - illustration on Fig3 on the score decomposition in the same fashion as PathSim (which counts meta-paths) ; notice the normalization step ### 3. The SemRec solution ##### 3.1 Basic idea - principle: evaluate similarity between users based on weighted and unweighted MP then infer scores from similar users - preference weights are given to different MP - difficulties: combine recommendations generated by different MP - pb1: important bias due to the fact that some types of paths are more numerous than others => similarity based on paths does not necessarily reflect similarity between objects => some kind of normalization to avoid that - pb2: we should personalize recommendations for better efficiency, but sparser data => recommendation by user groups with same preferences ##### 3.2 Recommendation with a single path - presentation of the method for one path (before generalizing) - supposing ratings from 1 to N - we have user similarity matrix for this type of specific path - compute Q_u,i,r: intensity of user u evaluating r item i from the similarity sum between users according to meta-path P_l - score predicted is the weighted average of ratings over Q_u,i,r ##### 3.3 Recommendation with multiple paths - now if we use several MP... - 3.3.1: compute weights to ratings corresponding to each type of path by minimizing mean squared error between actual and predicted scores - 3.3.2: personalized learning: each user as a weight vector, then same principle: minimizing MSE but with different weights with different users - 3.3.3: add a regularization process: learning difficult when we have few data => we use similar users ; regularization term to have the weight of a user similar to its neighbors average weight => eq 9 general form of the optimization goal ; optimization by projected gradient until convergence ##### 3.4 Discussion - general form of the objective function is L_3 in eq9 - if parameter lambda_1=0 => L_2 (equation 6) - if parameter lambda_1=infinity => L_1 (equation 4) - lambda_1 controls the level of personalization (the lower, the more personalized) - complexity analysis ### 4. Experiments ##### 4.1 Datasets - Douban (13400 u, 12700 i (= movies), 1M ratings de 1 à 5) - Yelp (16200 u, 14300 i (= buisnesses), 200K ratings de 1 à 5) - Douban clearly denser than Yelp - cf Tab.2 ##### 4.2 Metrics - accuracy evaluated with RMSE and MAE ##### 4.3 Comparison methods - 4 variants of their own model: SemRecReg (comprehensive), SemRecSgl (one type of MP), SemRecAll (same weight for everyone), SemRecInd (personalized weights, but no regularization (?)) - and methods from the literature: PMF (classic MF), SMF (PMF + reg), CMF (MF with HIN structure), HeteMF (MF + reg based on entity similarity) - no MP longer than 4 ##### 4.4 Effectiveness experiments - different settings training/test: 20/80, 40/60, 60/40, 80/20 for Douban ; 60/40, 70/30, 80/20, 90/10 for Yelp (because sparser) - Tab4: SemRecReg always have better performances in all conditions - other trends: SemRec with multiple paths in general better than SemRec with simple paths ; sparsity implies that SemRecInd is worse than SemRecAll in most circumstances (maybe I misunderstood something before) ; regularization has beneficial effects ##### 4.5 Study on cold start problem - cold start translated here by smaller training and larger test - Fig4: SemRecReg is clearly more performing in this context ##### 4.6 Study of weight preferences - explore the importance of weighted meta-paths (versus unweighted) - we can give some "slack" by imposing less constraints on scores (for example rating +/- 1) - Fig6: very demonstrative, perfs are clearly better when constraints are harder ### 5. Related work - ref 11 : HIN for data mining (Sun et Han, 2012) - ref 12 : similarity measures in HIN with PathSim (Sun et al. 2012) - ref 10 : similarity measures in HIN with HeteSim (Shi et al. 2014) - ref 2 : HIN for RS (Jamali and Lakshmanan, 2013) - the cold start question, difference depending on the technique used (MF, CF...) - ref 15 : closest to this paper, HeteRec (Yu et al., 2014) but do not use weighted paths

# In short Short article considering HIN-based recommendation. The contribution consists in using not only meta-paths in the HIN, but also "enhanced" meta-paths which are related to more elaborate motifs (typically triangles) # Summary ### 1. Introduction - usually HIN-based RS rely on the number pf meta-paths from u to i, the more there are, the higher the recommendation - Figure 1 summarizes situations that the authors want to distinguish: (u1,b4) and (u1,b3) are equivalent in terms of (P2) meta-paths, but not in terms of 3-nodes based "trust" patterns - patterns here are considered using 3 nodes patterns (à la Milo et al) as represented in Figure 2, meta-paths based on these motifs are called Motif Enhanced Meta Path (MEMP) - the concept is explored on 2 datasets: Epinions and CiaoDVD ### 2. Framework - to compute similarity using MP, we build the adjacency matrix W_{Ai,Aj} where Ai and Aj are types, then we build the commuting matrix Cp which is a product of the adjacency matrices corresponding to path P => similarities are based on counting obtained by matrix products ##### MoHINRec framework - they define adj matrix based on the same principle, but an edge is considered if it belongs to a pattern of interest (illustrated on Figure 3) - equation 2: they merge different matrices (obtained with usual MPs or MEMPs) with an alpha-weighting - then they test this with state-of-the-art HIN-based RS (actually, one they have designed in another work) ### 3. Experiments and analysis - 2 datasets, described in Table 1: - Epinions: ~22K users, 300K items, 900K ratings - CiaoDVD: ~17K users, 16K items, 72K ratings - evaluation metrics: MAE, RMSE (accuracy based) - baselines for comparison: RegSVD (MF with regularization), SoReg (MF using social links for regularization), SocialMF (MF with social trust propagation), FMG (IN-based RS, state-of-the-art) - supposedly better than the others - Settings: 8/1/1 train-validation-test, 5 experiment series with different splits ##### Performance comparison: - Summary in Table 2: MoHINRec outperforms the others - depending on the motif considered, we should vary the alpha coefficient to obtain the best possible performance (it is not clear for me how they tune alpha) - performances vary depending on the dataset and morif considered - FMG is still more efficient than othher benchmark methods ### 4. Related work ##### 4.1 HIN based recommendation ##### 4.2 motifs (very complex networks based: Milo et al., recent works by Leskovec et al.

Although certain NP-complete problems stay hard even on average input, some NP-complete problems are only hard in rare cases. Such NP-complete problems can provably be resolved very quickly on average case under some probability distributions over all possible inputs. **1. What is the average case of a real-world graph ? ** **2. How to properly define a suitable probability distribution for real-world graphs ? **

Summary: The paper compares several supervised learning techniques to detect money laundering in bitcoins, using a company labelled dataset that they provide publicly. This is a tutorial presented at a KDD 2019 workshop: https://sites.google.com/view/kdd-adf-2019 Strengths: I particularly loved the direct, clear and convincing writing style of the paper. Also, the authors (seem to) make their data publicly available, and it is a very interesting dataset (in particular thanks to the labelling). Weaknesses: The used data suffers from several limitations: it is very partial (approx 200 K transactions, while 400 M are available), and the labelling is not presented in details. More globally, several aspects of the work are not detailed, and many choices seem arbitrary. But the authors themselves state that this is "early experimental results" only, aimed at "inspiring others to work on this societally important challenge".