Text Matching Progress

Text matching is a core component in many natural language processing tasks, where many task can be viewed as a matching between two texts input.

\[f(s, t)=g(\psi(s), \phi(t), \eta(s, t))\]

Where $s$ and $t$ are source text input and target text input, respectively. The $\psi$ and $\phi$ are representation function for input $s$ and $t$, respectively. The $\eta$ is the interaction function, and $g$ is the aggregation function. More detailed explaination about this formula can be found on A Deep Look into Neural Ranking Models for Information Retrieval. The representative matching tasks are as follows:

Tasks Source Text Target Text
Ad-hoc Information Retrieval query document (title/content)
Community Question Answering question question/answer
Paraphrase Identification string1 string2
Natural Language Inference premise hypothesis
Response Retrieval context/utterances response

Updated: