Term position‐based language model for information retrieval

Abstract

Term position feature is widely and successfully used in IR and Web search engines, to enhance the retrieval effectiveness. This feature is essentially used for two purposes: to capture query terms proximity or to boost the weight of terms appearing in some parts of a document. In this paper, we are interested in this second category. We propose two novel query‐independent techniques based on absolute term positions in a document, whose goal is to boost the weight of terms appearing in the beginning of a document. The first one considers only the earliest occurrence of a term in a document. The second one takes into account all term positions in a document. We formalize each of these two techniques as a document model based on term position, and then we incorporate it into a basic language model (LM). Two smoothing techniques, Dirichlet and Jelinek‐Mercer, are considered in the basic LM. Experiments conducted on three TREC test collections show that our model, especially the version based on all term positions, achieves significant improvements over the baseline LMs, and it also often performs better than two state‐of‐the‐art baseline models, the chronological term rank model and the Markov random field model.

A retrieval model family based on the probability ranking principle for ad hoc retrieval

“…Therefore, it is necessary to evaluate the effect of going beyond such assumptions by experiments. Hence, possible topics of future study include combining PBR models with term association methods, such as cross terms (Zhao et al, 2014) or MRF techniques (Metzler & Croft, 2005) that incorporate various degrees of term dependencies, and the use of term position features (Hammache & Boughanem, 2021). Furthermore, we have ignored queryindependent features (by Assumption 3).…”

“…Apart from these models, some approaches that go beyond bag-of-words have been shown to be effective. These include the cross term approach that models term association (Zhao et al, 2014), term proximity matching methods such as Markov random field (MRF) (Metzler & Croft, 2005), and also the use of term position features (Hammache & Boughanem, 2021). As the objective of this paper is to demonstrate the effectiveness of the new bag-of-words PBR models in a pilot study, we restrict to bag-of-words baselines for a fair comparison.…”

Multistage BiCross encoder for multilingual access to COVID-19 health information

“…As the documents in the MLIA corpus are long, we only consider the first N sentences for inference, where N denotes average number of sentences in documents of the corpus. Moreover, previous research [ 29 , 30 ] shows that any relevant document is likely to contain relevant sentences at the beginning of the document. The document-level relevance score is determined by aggregating the top k scoring sentences in the document as follows: where BiScore(d) is the document-level relevance score for document d using the bi-encoder model.…”

Neural methods for effective, efficient, and exposure-aware information retrieval

“…For example, the challenge of vocabulary mismatch, and hence the importance of semantic matching, may be amplified when retrieving shorter text [95][96][97]. Similarly, when matching the query against longer text, it is informative to consider the positions of the matches [98][99][100], but may be less so in the case of short text matching. When specifically dealing with long text, the compute and memory requirements may be significantly higher for machine learned systems (e.g., [101]) and require careful design choices for mitigation.…”