Information retrieval finds documents that match a user’s query and ranks them by relevance. Web search, legal research, literature surveys, and related-article features all use this basic pattern. section 5.0 section 5.5
A retrieval system addresses two linked problems:
- Find documents matching the query. section 5.5
- Rank the matching documents according to their relevance to the user. section 5.5
- Represent each query-document pair with a relevance class such as least relevant, moderately relevant, or highly relevant. section 5.5
Legal-archive search
A paralegal searches previous cases to establish precedent and research case law; the system returns documents ordered by their relevance to the query. section 5.5
The retrieval task can be modeled with gradient-boosting classifiers. LightGBM supports the relevance-class case study, while XGBoost enables a later comparison and randomized cross-validation for parameter selection. section 5.5 section 6.5
