Information Quantity in Text and Its Applications

Human reading process significantly influences text understanding. Previous work has proposed a measure of information by simulating human reading process with a reading aim. The measure reflects both human memory of words in mind and association between words. There are two limitations: 1) interval between documents is an important reading factor and is not considered in the simulation, 2) the usefulness of the measure is limited to text recommendation in the work. This work proposes a multi-document scanning mechanism by exploiting the interval between documents and defines a measure named Information Quantity in Text in the mechanism. The measure is applied in both text recommendation and text summarization. Experiments show the measure outperforms an entropy-based baseline in determining the reading order of text sets according to the Summary Content Unit evaluation, and performs well in multi-document summarization according to the Pyramid evaluation. Experiments also show interval between documents in the scanning mechanism improves the recommendation and the summarization.