Using Latent Semantic Analysis and Word Matching to Enhance Bridging Reading Strategy Identification

The main goal of this study is to identify bridging reading strategy -a strategy that a reader uses to make a connection from the current sentence to previous sentences to help understanding the meaning of the text. For a specific target sentence, there are two types of bridging: local and distal. Benchmarks were created to help represent each type of bridging. The two immediate prior sentences of each target sentence together created a benchmark for the local bridging. The benchmarks for distal bridging were those prior sentences, excluding two immediate prior sentences. There were three ways that distal benchmarks were created: chunks based-on paragraph, chunks based-on target sentence, and entire collection of prior sentences. The results showed that using modified benchmark by removing up to 4 words within a threshold 0.4 has significantly improved the identification of distal bridging reading strategy by 14% from the original benchmark evaluation. On the other hand, to identify local bridging, using modified benchmark by removing 4 words has significantly improved the identification by 19% from the original benchmark evaluation.