Source-linked AI summary
A Survey on Personality-Aware Recommendation Systems
Sahraoui Dhelim, Nyothiri Aung, Mohammed Amine Bouras, Huansheng Ning, Erik Cambria
TL;DR
Personality-aware recommendation systems have proliferated, but their methods and applications vary across the literature. This survey systematically reviews and classifies these systems by personality modeling and recommendation technique, while covering datasets and challenges. It concludes that Big-Five models predominate and that personality-aware approaches are particularly advantageous for cold-start and data-sparsity problems, alongside privacy and measurement concerns.
Problem
The literature on personality-aware recommendation systems is rapidly expanding across different personality models, recommendation techniques, and applications, creating a need for systematic synthesis.
Method
The survey compares personality modeling methods and recommendation techniques, and reviews systems, datasets, benchmarks, and personality measurement approaches.
Results
The survey finds that most personality-aware recommendation systems use the Big-Five model and identifies advantages over conventional techniques for cold-start and data-sparsity problems.
Takeaways & Limitations
Personality-aware recommendation systems broaden recommendation design by incorporating personality information into measurement, matching, and filtering processes.
Abstract
from arXiv · showhide
With the emergence of personality computing as a new research field related to artificial intelligence and personality psychology, we have witnessed an unprecedented proliferation of personality-aware recommendation systems. Unlike conventional recommendation systems, these new systems solve traditional problems such as the cold start and data sparsity problems. This survey aims to study and systematically classify personality-aware recommendation systems. To the best of our knowledge, this survey is the first that focuses on personality-aware recommendation systems. We explore the different design choices of personality-aware recommendation systems, by comparing their personality modeling methods, as well as their recommendation techniques. Furthermore, we present the commonly used datasets and point out some of the challenges of personality-aware recommendation systems.
I. INTRODUCTION
Personality-aware recommendation systems extend conventional recommendation approaches by incorporating user personality information. This survey reviews their design choices, recommendation techniques, datasets, and open challenges.
- Personality-aware recommendation systems emerged from personality computing, which integrates personality psychology theories with computing systems.
- The survey addresses a rapidly proliferating research area and is presented as the first review focused specifically on personality-aware recommendation systems.
- The survey compares personality models and recommendation techniques while reviewing related systems, datasets, benchmarks, and research organization.
- Conventional systems typically perform rating, filtering, and recommendation, using collaborative, content-based, or hybrid filtering techniques.
- Personality-aware systems add personality measurement and personality matching phases before rating, then incorporate personality information into filtering similarity measurements.
A. Personality measurement
Personality measurement is central to personality-aware recommendation because misidentifying a user’s personality can reduce recommendation accuracy. Systems use questionnaires or automatic recognition, balancing questionnaire accuracy against user burden and practical feasibility.
- Personality measurement is the most important phase because misidentifying a user’s personality can negatively influence recommendation accuracy.
- Questionnaire-based measurement is generally more accurate than automatic personality recognition, but automatic recognition is easier when existing user data are available.
- Self-report questionnaires commonly assess personality differences using five-level Likert-scale responses.
- Long questionnaires include NEO-FFI, NEO-PI-R, IPIP-NEO, and BFI, whereas personality-aware systems prefer shorter 5–10-item instruments such as BFI-10 and TIPI.Shorter questionnaires reduce the time required from users during registration.
- Questionnaires face self-bias and reference-group effects, although the survey states self-bias does not affect personality-aware recommendation systems because users have no benefit from misleading the system.
2) Automatic personality recognition:
Automatic personality recognition maps user-related data to personality scores when questionnaires are unavailable or inconvenient. The survey distinguishes text-, multimedia-, and behavior-based approaches, with text-based recognition generally reported as more accurate.
- Automatic personality recognition maps subject-related data to a personality score representing the subject’s personality type.
- The three main automatic-recognition classes are text-based, multimedia-based, and behavior-based approaches.Their inputs include text, images or audiovisual data, and behavioral patterns such as gaming or browsing.
- Text-based automatic personality recognition generally has higher accuracy than multimedia-based and behavior-based recognition.
- Text-based recognition infers personality traits from word choices in social-network posts or messages, commonly using Linguistic Inquiry and Word Count.
C. Personality filtering
Personality filtering adapts collaborative filtering by forming neighborhoods using personality similarity or combined personality-rating similarity. The survey distinguishes personality-neighborhood methods from matrix factorization and identifies several designs for computing similarity and predictions.
- Personality-aware filtering forms user neighborhoods from personality-trait similarity or hybrid personality-rating similarity.
- Filtering methods comprise personality filtering, personality matching, and hybrid filtering; personality filtering further divides into neighborhood and matrix factorization methods.
- Personality-neighborhood methods: Personality-neighborhood methods commonly use Pearson correlation to measure rating similarity before predicting ratings from neighboring users.
- Personality-neighborhood methods: Personality similarity can be computed with alternative proximity functions rather than a single standard measure.
- Personality-neighborhood methods: Prediction scores aggregate neighborhood ratings and similarities, using neighbors who previously rated the target item.
2) Matrix factorization methods:
Personality-enhanced matrix factorization extends conventional latent-factor recommendation by incorporating users’ personality traits and, in some systems, other social attributes.
- Matrix factorization decomposes user-item interactions into low-dimensional user and item latent vectors.
- The conventional preference estimate is the dot product of a user latent vector and an item latent vector.
- Personality-enhanced matrix factorization adds personality traits as additional latent attributes alongside ratings, sometimes including gender or age group.
B. Personality matching methods
Personality matching assigns personality scores to recommended items and matches them with target users, using content, attribute, or hybrid analysis. This approach can avoid neighborhood computation and mitigate cold-start effects when matching criteria are observable.
- Personality matching assigns each item a personality score through content analysis, attribute analysis, or a hybrid approach.
- Content analysis can infer an item’s personality from textual descriptions, labels, categories, or other item content.
- Personality matching is applied when recommended content and target users share observable matching criteria.
- Because matching does not require users’ ratings, it can alleviate cold-start effects and avoid extensive neighborhood computation.
C. Hybrid personality filtering methods
Hybrid personality filtering combines personality-based user filtering with personality matching in the item space. The survey also places deep-learning methods within the broader classification of personality-aware recommendation techniques.
- Hybrid personality filtering methods: Hybrid personality filtering combines user-space personality filtering with item-space personality matching.
- Hybrid personality filtering methods: Hybrid methods leverage personality filtering to identify users with similar ratings and personality matching to evaluate recommended items.
- Deep learning methods: Deep learning models are used either to detect user personality or directly within the recommendation process.
- Deep learning methods: The survey classifies personality-aware recommendation systems by recommendation technique and summarizes techniques in its classification tables.
IV. LITERATURE REVIEW OF PERSONALITY-AWARE RECOMMENDATION SYSTEMS
Personality-aware recommendation has been applied across social-network, television, movie, and friend recommendation settings. These systems use personality traits alongside conventional signals to support prediction, personalization, and cold-start recommendation.
- PersoNet uses Big-Five traits in hybrid friend selection and outperforms conventional rating-based hybrid filtering, including acceptable precision and recall during cold start.
- Followee prediction studies incorporate personality traits with state-of-the-art factors and evaluate content-based prediction using a Twitter dataset of 1852 English-language users.
- ROPPSA uses normalization and folksonomy procedures to generate group TV recommendations for viewers with similar personality traits.
- HyPeRM combines personality character with demographic information in a hybrid movie recommender, while another system adapts its algorithm to users’ character criteria.
- Studies link personality traits with movie preferences and investigate inferring personality from implicit movie behaviors to support movie recommendation.
C. Music recommendations
Personality-aware music and image recommendation research connects personality traits with preferences, behavior, content features, and recommendation choices. Academic recommendation systems also use personality and interpersonal information to support conference participation and collaboration.
- Music recommendations: Hybrid music recommendation matches users’ personality traits with features extracted from song audio.
- Music recommendations: Personality-aware music models examine relationships between personality, listener behavior, preferences, and needs, while inferring traits from social-media data.
- Music recommendations: A general model deduces music preferences from personality characteristics, with subject studies used to evaluate the approach.
- Image recommendations: Image features can help infer user personality, while personality traits can support relevant image recommendation; Instagram research found agreeableness most associated with photo and color features.
- Academic recommendations: SPARP recommends academic conference participants using a hybrid model incorporating interpersonal relationships and personality information.
F. Product recommendations
Personality-aware recommendation extends to products, games, points of interest, and tourism. Systems combine personality with interests, language-derived lifestyles, similarity, trust, content, and contextual factors to tailor recommendations.
- Product recommendations: Meta-Interest combines user interests and personality traits with user-item metapaths for product recommendation.
- Product recommendations: Big-Five personality traits provide a basis for measuring user similarity in collaborative filtering for product recommendation.
- Product recommendations: A data-driven product recommender derives seven lifestyle dimensions from shoppers’ language usage and uses them to predict buying preferences.
- Game recommendations: A game recommender extracts players’ personality types from social-network posts and associates game content with personality types, evaluating the system on 63 players and more than 2000 games.
- POI and tourism recommendations: A trust-based POI recommender uses personality similarity to compute trust alongside graphic and temporal influence, while tourism systems recommend types compatible with personality.
V. DATASETS AND BENCHMARKS
Personality-aware recommendation research relies on public datasets containing personality information, ratings, content interactions, and user preferences. The survey highlights myPersonality, MovieLens, and Newsfullness as widely used resources.
- Datasets and benchmarks: Open public datasets containing user personality information enable training and comparison of personality-aware recommendation models against state-of-the-art benchmarks.
- myPersonality: myPersonality is a Facebook application based on the IPIP version of the NEO personality inventory that displays users’ personality scores.
- MovieLens: MovieLens provides movie-rating data collected over multiple periods and dataset versions, with MovieLens 25M containing 25 million ratings and one million tag applications for 62,000 movies.
- Newsfullness: Newsfullness contains TIPI scores for 2228 users and records the news articles they viewed or liked across major websites and diverse categories.
VI. CHALLENGES AND OPEN ISSUES
Personality-aware recommendation systems can address cold start and data sparsity, but their benefits introduce privacy and personality-detection accuracy challenges. These challenges matter because inaccurate personality measurements can produce inaccurate recommendations, while personality data requires protection.
- A. Personality information privacy: Personality information is especially sensitive, and systems must balance data sharing with protecting users from personality inference through recommendation results.The survey highlights risks including malicious analysis of recommendations to predict users’ personality types.
- B. Measurement accuracy: Inaccurate personality measurement inevitably leads to inaccurate recommendations, and treating traits as infrequently changing can prolong those inaccuracies.Questionnaires may be affected by reference-group effects, while automatic recognition can also misidentify personality traits.
- B. Measurement accuracy: Image-based automatic personality recognition can misinterpret shared photos when it ignores their context, such as occupationally motivated photography.Examples include photographers sharing nature images and journalists sharing war photos that may not reflect their personalities.
- Personality-aware recommendation systems address recommendation challenges such as cold start and data sparsity, but introduce privacy and personality-detection accuracy challenges.The survey identifies these benefits alongside the need to preserve personality information and maintain accurate detection.