Page 181 - tech fest 2025 ב״ש חוברת תקצירים

Book of Abstracts | 2025 A Support Application for War Victims and Individuals Facing Emotional Challenges
SE-D-05
Avital Chikota; avitach@ac.sce.ac.il Eden Maimoni; edenma6@ac.sce.ac.il
Advisor: Dr. Karim Abu-Affash
SCE - Shamoon College of Engineering, Be'er-Sheva
We present an application to provide social support for war victims and individuals struggling with anxiety, depression, and post-traumatic stress disorder (PTSD). It integrates an AI-powered chat system that offers immediate emotional assistance and allows users to share personal experiences anonymously and engage with a supportive community. The content platform enables users to publish posts, comment, search, and apply advanced filtering options. A location-based feature facilitates participation in social events and nearby gatherings using GPS. It provides rapid access to essential information and professional services, including an emergency button in distress situations. Users can also gain access to information on legal rights and complete self-assessment questionnaires. Overall, it offers an intuitive user experience and fosters community-driven support tailored to the digital age.
Keywords: ai, anonymously, emergency button, essential information, gps, supportive community, user experience
The Impact of Code Summarization on Code Retrieval Performance
SE-D-06
Nir Kanimach; nironi1998@gmail.com Lior Gofman; gofmanlior@gmail.com
Advisor: Dr. Marina Litvak
SCE - Shamoon College of Engineering, Be’er-Sheva
This project explores a novel code retrieval method to support software reuse. Rather than retrieving code directly through queries, our approach uses large language models (LLMs) to generate natural language summaries of code snippets and match them to queries. We assess whether summary-based retrieval outperforms traditional approaches using code or documentation. Based on the CodeSearchNet dataset, we tested multiple LLMs (LLaMA, GPT, CodeT5, DeepSeek) for summarization. Each model produced a dataset of code samples, queries, and corresponding summaries. To measure retrieval quality, we used the CoIR competition model, a strong baseline from the Code Search track of the NeurIPS CoIR 2022 challenge. We report comparative evaluation results showing DeepSeek as the best-performing model and the best CoIR competition system.
Keywords: code retrieval, code summarization, codesearchnet, coir, deep learning, large language models, software reuse
181