CS MSc Thesis Presentation 27 August 2026

Thursday, 27 August there will be a master thesis presentation in Computer Science at Lund University, Faculty of Engineering.

The presentation will take place in E:4130 (Lucas).

Note to potential opponents: Register as an opponent to the presentation of your choice by sending an email to the examiner for that presentation (firstname [dot] lastname [at] cs [dot] lth [dot] se). Do not forget to specify the presentation you register for! Note that the number of opponents may be limited (often to two), so you might be forced to choose another presentation if you register too late. Registrations are individual, just as the oppositions are! More instructions for opponents are found here on the LTH thesis project page.

13:15-14:00 in E:4130 (Lucas)

• Presenters: Jesper Follin, David Bergkvist
• Title: Intelligent E-commerce Product Profile Generation
• Examiner: Sule Tekkesinoglu
• Supervisor: Elizabeth Bjarnason (LTH)

E-commerce product profiles consists of product information that influences customers' purchasing decisions and business performance. Manual creation requires collecting, interpreting and transcribing information from heterogeneous sources, making it laborious, error-prone, and difficult to scale. Through an exploratory case study with a Swedish home-improvement e-commerce retailer, this thesis identifies workflow challenges, derives automation requirements, and develops a vendor-agnostic pipeline that uses large language models to generate structured Swedish product profiles. A blind human evaluation compares Gemini-3.1-Pro and Gemma-4-31B, with pages represented as trimmed HTML or Markdown. Overall profile quality varied little across configurations. Markdown reduced product page token counts by a median of 76\%. Gemma 4 with Markdown had 99\% lower inference costs than Gemini 3.1 Pro with trimmed HTML. Direct information extraction was generally reliable, while tasks involving selection, rewriting or judgement were more prone to product information failures. The findings support automated product profile generation followed by human review.