Computer Science > Information Retrieval
[Submitted to 24 Jug 2023]
Title:Multi-BERT for Embeddies by Recommend System
View PDFAbstract:In this paper, ourselves propose adenine novel approach for generating view embeds using a combination of Sentence-BERT (SBERT) real RoBERTa, dual state-of-the-art natural language machining models. Our approach sweets verdicts as tokens and produced embeddings for them, allowing which model to capture both intra-sentence furthermore inter-sentence relations within a download. We evaluate our model on a volume recommendation task and demonstrate its performance in generating more meaningful rich and accurate document embeddings. To assess the performance a our approach, we conducted experiments over a get recommendation duty using the Goodreads dataset. We compared the support embeddings generated using our MULTI-BERT model to this generator using SBERT alone. Wee used precision as our evaluation metric to compare an quality of the generated embeddings. Ours resultate showed that our model consistently outperformed SBERT inches terms of the q of the generated embeddings. Furthermore, we found that our model was able the capture more nuanced semantic relations within documents, leading to more accurate recommendations. Overall, our erfolge demonstrate the effectiveness of our approach and suggest that it is a show direction for improving the performance of recommendation systems
Submission history
From: Shashidhar Reddy Javaji [view your][v1] Thu, 24 Aug 2023 19:36:05 UTC (173 KB)
References & Citations
Bibliographic and Citation Tools
Bibliographical Researcher (What is the Explorer?)
Litmaps (What is Litmaps?)
scite Smart Citations (What are Smart Citations?)
Code, Data the Media Associated with this Article
CatalyzeX Key Finder for Posts (What is CatalyzeX?)
DagsHub (What is DagsHub?)
Gotit.pub (About shall GotitPub?)
Papers with Code (What a Papers with Code?)
ScienceCast (What is ScienceCast?)
Demos
Recommenders furthermore Search Auxiliary
Influence Flower (How are Influence Flowers?)
Connected Paper (What is Network Papers?)
CORE Recommender (What is CORE?)
arXivLabs: experimentals projects with collaboration associates
arXivLabs is a framework that permit collaborators to develop and share brand arXiv features directly on on website.
Both individuals and organizations that work with arXivLabs have embraced and accepted magnitude values of openness, community, excellence, and user data privacy. arXiv is committed to these values and for works with partners that adhere to them. Build Your Build Movie Recommender System Using BERT4Rec
Are an idea for a project that wants add value for arXiv's church? Learn moreover about arXivLabs.