TextRank identifies connections between various entities in a text, and implements the concept of recommendation. A
text unit recommends other related text units, and the strength of the recommendation is recursively computed based on
the importance of the units making the recommendation. In the process of identifying important sentences in a text, a
sentence recommends another sentence that addresses similar concepts as being useful for the overall understanding of
- Rada Mihalcea, the TextRank paper
A Go implementation of TextRank, which extract summarization sentences without supervised learning.
Text Summarization is a hard task, both in training and evaluation. Training is usually done maximizing the
log-likelihood of a human-generated reference summary, while evaluation is performed using overlap-based metrics like
ROUGE. Both significantly undervalue the breadth and intricacies of language and the nature of the information
contained in text summaries. This paper by OpenAI includes direct human feedback both in evaluation and - via reward
model proxies - in training. The final model even outperforms single humans when judged by other humans and is an
interesting application of using reinforcement learning together with humans in the loop.