Rule-based summarization extracts important text with predefined linguistic rules and heuristics rather than adapting through machine learning. Sentence position, length, keyword frequency, and cue phrases can guide selection. The approach is straightforward and interpretable, but rigidity can make it miss the nuances of different texts. section 6.3.4
Common rule-based approaches include:
- Sentence scoring and extraction ranks sentences using criteria such as position, keywords, or TF-IDF, then combines the highest-scoring sentences. section 6.3.4
- TextRank represents sentences as graph nodes, connects overlapping content, and extracts central sentences. section 6.3.4
- Frequency-based methods extract frequent keywords, n-grams, or phrases that represent major topics. section 6.3.4
Customer-review summaries
For “I received it well before the stipulated deadline,” frequency-based extraction produced “received,” “well,” and “before,” while a generated alternative captured the idea as “Early delivery.” Such outputs show both the transparency and the occasional thinness of rule-based summaries.
