Cognitive-aware Short-text Understanding for Inferring Professions

TL;DR

This paper introduces a novel cognitive-aware framework for inferring authors' professions from short-texts, effectively handling lexical noise and latent semantic features, and combining multiple algorithms for improved accuracy.

Contribution

The paper proposes a multi-aspect cognitive feature-based model that integrates curve fitting, support vector, and boosting algorithms for occupation inference from short-texts.

Findings

Outperforms existing methods in Twitter-based occupation prediction

Achieves higher accuracy by leveraging cognitive and contextual features

Demonstrates robustness against lexical noise in short-texts

Abstract

Leveraging short-text contents to estimate the occupation of microblog authors has significant gains in many applications. Yet challenges abound. Firstly brief textual contents come with excessive lexical noise that makes the inference problem challenging. Secondly, cognitive-semantics are not evident, and important linguistic features are latent in short-text contents. Thirdly, it is hard to measure the correlation between the cognitive short-text semantics and the features pertaining various occupations. We argue that the multi-aspect cognitive features are needed to correctly associate short-text contents to a particular job and discover suitable people for the careers. To this end, we devise a novel framework that on the one hand, can infer short-text contents and exploit cognitive features, and on the other hand, fuses various adopted novel algorithms, such as curve fitting, support vector, and boosting modules to better predict the occupation of the authors. The final estimation module manufactures the $R^w$-tree via coherence weight to tune the best outcome in the inferring process. We conduct comprehensive experiments on real-life Twitter data. The experimental results show that compared to other rivals, our cognitive multi-aspect model can achieve a higher performance in the career estimation procedure, where it is inevitable to neglect the contextual semantics of users.