A New Digital Signal Processing Based Model with Multi-Aspect Term Frequency for Information Retrieval

In recent decades, the vast majority of researchers in the field of information retrieval (IR) have been studying three main categories of IR models (i.e., vector space models, probabilistic models and statistical language models). Recently, some researchers have been exploring a new category of IR models which introduce the knowledge from the field of digital signal processing (DSP), which have been shown to be promising. However, the existing DSP-based models are not well-performed in some cases because they have not incorporated effective term weighting methods and the existing framework itself has some disadvantages to be overcome. In our research, we propose a new DSP-based IR model, denoted as DSP-MATF, which incorporates a very effective term weighing method from the well-performed Vector Space Model named Multi-Aspect Term Frequency (MATF). In addition, for improving the existing DSP-based IR frameworks, we consider each query term as a spectrum enveloped by different curves of seven kernel functions. To testify the effectiveness of our proposed model, we conduct extensive experiments on seven standard TREC datasets. The results show that in most cases our proposed model outperforms the strong baselines in terms of varied metrics.