Exploring Explanations for Matrix Factorization Recommender Systems

Rastegarpanah, Bashir; Crovella, Mark; Gummadi, Krishna P.

doi:10.18122/b2r717

Cited by 8 publications

(8 citation statements)

References 3 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Explanations can also be used to justify or describe [80]. While early recommender system explanation approaches provided a uniform explanation style for single-source collaborative filtering [30], more recent work explores how to derive explanations for recommender systems based on 'hybrid' multiple sources [36] and matrix factorisation [61]. However, a reported need for explanation may not always correspond to differences in behaviour or performance; in a study of news recommender systems, end-users expressed a desire for explanations, but the number of news items they opened did not change when provided with reasons for their recommendations [70].…”

Section: Interpreting Intelligent Systemsmentioning

confidence: 99%

'It's Reducing a Human Being to a Percentage'

Binns

Kleek

Veale

et al. 2018

Proceedings of the 2018 CHI Conference on Human Factors in Computing Systems

294

View full text Add to dashboard Cite

Data-driven decision-making consequential to individuals raises important questions of accountability and justice. Indeed, European law provides individuals limited rights to 'meaningful information about the logic' behind significant, autonomous decisions such as loan approvals, insurance quotes, and CV filtering. We undertake three experimental studies examining people's perceptions of justice in algorithmic decision-making under different scenarios and explanation styles. Dimensions of justice previously observed in response to human decision-making appear similarly engaged in response to algorithmic decisions. Qualitative analysis identified several concerns and heuristics involved in justice perceptions including arbitrariness, generalisation, and (in)dignity. Quantitative analysis indicates that explanation styles primarily matter to justice perceptions only when subjects are exposed to multiple different styles-under repeated exposure of one style, scenario effects obscure any explanation effects. Our results suggests there may be no 'best' approach to explaining algorithmic decisions, and that reflection on their automated nature both implicates and mitigates justice dimensions.

show abstract

Section: Interpreting Intelligent Systemsmentioning

confidence: 99%

'It's Reducing a Human Being to a Percentage'

Binns

Kleek

Veale

et al. 2018

Proceedings of the 2018 CHI Conference on Human Factors in Computing Systems

294

View full text Add to dashboard Cite

show abstract

“…Because there is usually a tradeoff between explainability and recommendation accuracy, some research has focused on post-hoc explanainability of powerful black-box models. Such work includes ( Rastegarpanah et al, 2017 ) which explains MF-based recommender systems using influence functions to determine the effect of each user rating on the recommendation. Cheng et al (2019) also uses an influence-based approach to measure the impact of user-item interactions on a prediction and provides neighborhood-style explanations.…”

Section: Related Workmentioning

confidence: 99%

Sequence-Based Explainable Hybrid Song Recommendation

2021

View full text Add to dashboard Cite

Despite advances in deep learning methods for song recommendation, most existing methods do not take advantage of the sequential nature of song content. In addition, there is a lack of methods that can explain their predictions using the content of recommended songs and only a few approaches can handle the item cold start problem. In this work, we propose a hybrid deep learning model that uses collaborative filtering (CF) and deep learning sequence models on the Musical Instrument Digital Interface (MIDI) content of songs to provide accurate recommendations, while also being able to generate a relevant, personalized explanation for each recommended song. Compared to state-of-the-art methods, our validation experiments showed that in addition to generating explainable recommendations, our model stood out among the top performers in terms of recommendation accuracy and the ability to handle the item cold start problem. Moreover, validation shows that our personalized explanations capture properties that are in accordance with the user’s preferences.

show abstract

“…{1, 2, 3, 4, 5}, when applying the item-based approach. Equation (12) details how to compute the probability that the item i 1 be rated with 1 according to equation 3.…”

Section: E Running Examplementioning

confidence: 99%

“…In the context of RS, the main problem of matrix factorization is that the learnt latent space is not easy to interpret [11], so these models are not amenable to explaining their results [12]. The proposed model in [10] has been developed in order to alleviate this problem by applying a probabilistic approach for interpreting the factors of users and items.…”

Section: Introductionmentioning

confidence: 99%

A Collaborative Filtering Approach Based on Naïve Bayes Classifier

Valdiviezo-Díaz

Ortega

Cobos³

et al. 2019

IEEE Access

View full text Add to dashboard Cite

Recommender system is an information filtering tool used to alleviate information overload for users on the web. Collaborative filtering recommends items to users based on their historical rating information. There are two approaches: memory-based, which usually provides inaccurate but explainable recommendations; and model-based, whose recommendations are more precise but hard to understand. Here we propose a Bayesian model that not only provides us with recommendations as good as matrix factorization models, but these predictions can also be explained. The model is based on both user-based and item-based collaborative filtering approaches, which recommends items by using similar users' and items' information, respectively. Experiments carried out using four datasets present good results compared to several state-ofthe-art baselines, achieving the best performance using the Normalized Discounted Cumulative Gain (nDCG) quality measure and also improving the prediction's accuracy in some datasets.

show abstract

Exploring Explanations for Matrix Factorization Recommender Systems

Cited by 8 publications

References 3 publications

'It's Reducing a Human Being to a Percentage'

'It's Reducing a Human Being to a Percentage'

Sequence-Based Explainable Hybrid Song Recommendation

A Collaborative Filtering Approach Based on Naïve Bayes Classifier

Contact Info

Product

Resources

About