# From Scattered Sources to Comprehensive Technology Landscape: A Recommendation-based Retrieval Approach

**Authors:** Chi Thang Duong, Dimitri Percia David, Ljiljana Dolamic, Alain Mermoud, Vincent Lenders, Karl Aberer  
**arXiv:** [2112.04810](https://arxiv.org/abs/2112.04810)  
**Date:** 2021-12-09  
**Summary page:** https://www.summarizepaper.com/en/paper/2112.04810v1/

## Key points

- End-to-end framework for extracting and classifying technological mentions from company corpuses
- Technology classifier based on DistilBERT model with fine-tuning for better accuracy
- Recommendation-based retrieval model for more relevant results
- Comparison with tf-idf based retrieval method, showing significant outperformance in both company-company and technology-company retrieval tasks
- Promise of the proposed framework in automating technology and company retrieval from raw web data by leveraging advanced language models and recommendation-based techniques

## Summary

, , , , 

The paper presents an end-to-end framework for extracting and classifying technological mentions from company corpuses, as well as retrieving related technologies and companies. The technology classifier is based on the DistilBERT model with fine-tuning and refinement to achieve better accuracy. The recommendation-based retrieval model enables more relevant results. The authors also discuss existing works in technology monitoring and forecasting, specifically in the context of technology landscape monitoring. They mention various methods such as keyword-based or entity-based approaches for technology retrieval, citing examples like automated frameworks for detecting new technologies in texts, measuring proximity between patents and a company's technological footprint, and developing models for technology forecasting based on text mining techniques. Furthermore, the authors compare their approach with a tf-idf based retrieval method using tf-idf as a feature for technology classification and retrieval. Their approach significantly outperforms the baseline in both company-company retrieval and technology-company retrieval tasks. The improved performance is attributed to their better technology classifier utilizing a language model and their recommendation retrieval model considering relationships between companies, technologies, and their similarities simultaneously. Overall, the proposed framework shows promise in automating the retrieval of technologies and associated companies from raw web data by leveraging advanced language models and recommendation-based techniques. Future works may include allowing users to reformulate queries for better capturing intentions or combining technology classification with entity extraction for more accurate results.