Psycholinguistic Tripartite Graph Network for Personality Detection

Abstract

Most of the recent work on personality detection from online posts adopts multifarious deep neural networks to represent the posts and builds predictive models in a data-driven manner, without the exploitation of psycholinguistic knowledge that may unveil the connections between one's language usage and his psychological traits. In this paper, we propose a psycholinguistic knowledge-based tripartite graph network, TrigNet, which consists of a tripartite graph network and a BERT-based graph initializer. The graph network injects structural psycholinguistic knowledge from LIWC, a computerized instrument for psycholinguistic analysis, by constructing a heterogeneous tripartite graph. The graph initializer is employed to provide initial embeddings for the graph nodes. To reduce the computational cost in graph learning, we further propose a novel flow graph attention network (GAT) that only transmits messages between neighboring parties in the tripartite graph. Benefiting from the tripartite graph, TrigNet can aggregate post information from a psychological perspective, which is a novel way of exploiting domain knowledge. Extensive experiments on two datasets show that TrigNet outperforms the existing state-of-art model by 3.47 and 2.10 points in average F1. Moreover, the flow GAT reduces the FLOPS and Memory measures by 38% and 32%, respectively, in comparison to the original GAT in our setting.

LLM vs Small Model? Large Language Model Based Text Augmentation Enhanced Personality Detection Model

“…Personality detection can be defined as a multi-document multi-label classification task (Yang et al 2021b(Yang et al , 2023. Formally, given a set P = {p 1 , p 2 .…”

“…For example, Transformer-MD (Yang et al 2021a) stores posts' hidden states in the memory of Transformer-XL (Dai et al 2019) in order to avoid introducing post-order bias. TrigNet (Yang et al 2021b) constructs a heterogeneous graph between posts for each user based on the psycho-linguistic knowledge in LIWC and aggregates useful information with a GAT. D-DGCN (Yang et al 2023) builds a dynamic graph, which enables the model to learn the connections between the posts, and employs DGCN to integrate the information.…”

“…TrigNet (Yang et al 2021b): TrigNet constructs a trigraph for each user consisting of posts, words, and word categories, based on LIWC dictionary. It then employs a modified GAT to encode the graph and uses average pooling to obtain a user representation.…”

“…One main challenge is the scarcity of ground-truth personality traits, as they are collected from self-report questionnaires, which is time-consuming and often raises concerns related to user privacy. To overcome the issue, pre-trained language models have been applied to learn post representations (Yang et al 2021b;Xue et al 2018;Yang et al 2021b). To further improve the post representations, TrigNet (Yang et al 2021b) considers the contextual information of a user's posts as well as the external psycho-linguistic knowledge from LIWC (Xue et al 2018).…”

“…To overcome the issue, pre-trained language models have been applied to learn post representations (Yang et al 2021b;Xue et al 2018;Yang et al 2021b). To further improve the post representations, TrigNet (Yang et al 2021b) considers the contextual information of a user's posts as well as the external psycho-linguistic knowledge from LIWC (Xue et al 2018). However, the learned post representations are still unsatisfactory, resulting in inferior performance.…”