CR-SoRec: BERT driven Consistency Regularization for Social Recommendation Tushar Prakash ∗ tushar121prakash@gmail.com Sony Research India Bangalore, India Raksha Jalan ∗ rakshajalan9@gmail.com Sony Research India Bangalore, India Brijraj Singh Sony Research India Bangalore, India Naoyuki Onoe Sony Tokyo, Japan ABSTRACT In the real world, when we seek our friends’ opinions on various items or events, we request verbal social recommendations. It has been observed that we often turn to our friends for recommenda- tions on a daily basis. The emergence of online social platforms has enabled users to share their opinion with their social connec- tions. Therefore, we should consider users’ social connections to enhance online recommendation performance. The social recom- mendation aims to fuse social links with user-item interactions to ofer more relevant recommendations. Several eforts have been made to develop an efective social recommendation system. How- ever, there are two signifcant limitations to current methods: First, they haven’t thoroughly explored the intricate relationships be- tween the diverse infuences of neighbours on users’ preferences. Second, existing models are vulnerable to overftting due to the relatively low number of user-item interaction records in the inter- action space. For the aforementioned problems, this paper ofers a novel framework called CR-SoRec, an efective recommendation model based on BERT and consistency regularization. This model incorporates Bidirectional Encoder Representations from Trans- former(BERT) to learn bidirectional context-aware user and item embeddings with neighbourhood sampling. The neighbourhood Sampling technique samples the most infuential neighbours for all the users/ items. Further, to efectively use the available user-item interaction data and social ties, we leverage diverse perspectives via consistency regularization to harness the underlying informa- tion. The main objective of our model is to predict the next item that a user would interact with based on its interaction behaviour and social connections. Experimental results show that our model defnes a new state-of-the-art on various datasets and outperforms previous work by a signifcant margin. Extensive experiments are also conducted to analyze the proposed method. ∗ Both authors contributed equally to this research. Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for proft or commercial advantage and that copies bear this notice and the full citation on the frst page. Copyrights for components of this work owned by others than the author(s) must be honored. Abstracting with credit is permitted. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specifc permission and/or a fee. Request permissions from permissions@acm.org. RecSys ’23, September 18ś22, 2023, Singapore, Singapore © 2023 Copyright held by the owner/author(s). Publication rights licensed to ACM. ACM ISBN 979-8-4007-0241-9/23/09. . . $15.00 https://doi.org/10.1145/3604915.3608844 CCS CONCEPTS · Information systems→Social Recommendation;· Comput- ing methodologies→Neural networks; KEYWORDS Social Recommendation, BERT , Consistency Regularization ACM Reference Format: Tushar Prakash, Raksha Jalan, Brijraj Singh, and Naoyuki Onoe. 2023. CR- SoRec: BERT driven Consistency Regularization for Social Recommendation. In Seventeenth ACM Conference on Recommender Systems (RecSys ’23), Sep- tember 18ś22, 2023, Singapore, Singapore. ACM, New York, NY, USA, 7 pages. https://doi.org/10.1145/3604915.3608844 1 INTRODUCTION Numerous e-commerce websites and online platforms have evolved into popular social platforms as social media has grown in popular- ity to improve user engagement. Amazon’s OTT platform, Prime Video has introduced "Watch Party", encouraging users to invite friends to watch content simultaneously and socialise virtually. Similarly, a popular music-streaming platform, Spotify, has a fea- ture, "Blend", that allows users to invite friends and share their playlists. With the increasing popularity of such platforms, devel- oping recommendation systems that embrace social interactions into the recommendation model is essential. Social relationships among users may reveal their diverse interest trends and can be utilised for modelling user preferences more accurately. However, the complexity of high-order social relations makes it challenging to extract the most relevant data for modelling user preferences. Existing approaches fail to anticipate the multifaceted relationship between the diverse infuences of users’ neighbours on their pref- erences. For the recommendation model to be trained efciently, there must be an abundance of user-item interactions. However, user-item interaction data are extremely sparse in the interaction space. As a result, models trained on these data are vulnerable to over-ftting. To address the above mentioned problems, we propose a novel framework -BERT driven Consistency Regularization for Social Rec- ommendation (CR-SoRec). It learns efcient user-item interaction and user-user social representations by leveraging BERT along with Consistency Regularization Framework. In this paper, we have pro- posed an innovative way to generate robust user-item interactions representation (or social links representation ) by utilizing user header with neighbourhood sampling. Neighbourhood sampling is 883