A Measurement-Based Quantum-Like Language Model for Text Matching.

Quantum language models have attracted extensive attention in natural language processing tasks. Nevertheless, in this field measurement operators are mostly generated by randomly initialized parameter matrices, which cannot well explain the role of measurement operators in quantum theory. In this paper, we propose a Measurement-Based Quantum-like Language Model (MBQLM). Specifically, each word is considered a fundamental event in quantum probability space, which is a quantum state represented by a density matrix. We take the word density matrix in one sentence as a set of measurement operators to measure another sentence, which is consistent with the definition of measurement operators in quantum theory and has a specific semantic interpretation. The measured sentence state matrix representation effectively interact the feature information between sentences. To evaluate the performance of our model, we conduct experiments on WikiQA and TREC-QA datasets. The results show that our model achieves better performance than all benchmarks.