Summary of A Comparative Analysis Of Embedding Models For Patent Similarity, by Grazia Sveva Ascione and Valerio Sterzi
A comparative analysis of embedding models for patent similarity
by Grazia Sveva Ascione, Valerio Sterzi
First submitted to arxiv on: 25 Mar 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Information Retrieval (cs.IR); Machine Learning (cs.LG)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The paper presents two contributions to the field of text-based patent similarity. It compares different types of patent-specific pretrained embedding models, including static word embeddings and contextual word embeddings, on the task of calculating patent similarity. Specifically, it evaluates the performance of Sentence Transformers (SBERT) architectures with varying training phases on this task. The paper uses information about patent interferences as a proxy for maximum similarity between two patents to evaluate the models’ performance. The results show that Patent SBERT-adapt-ub outperforms the current state-of-the-art in patent similarity, and that large static models can still perform comparably to contextual ones when trained on extensive data. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper compares different types of models for calculating patent similarity. It looks at how well they do on a task called patent similarity calculation. The authors use information about when two patents are proven to be overlapping as a way to measure how good the models are. They find that one model, Patent SBERT-adapt-ub, is better than others at doing this job. They also find that bigger static models can still do well even if they’re not using contextual information. |
Keywords
* Artificial intelligence * Embedding
