Word-Level Contextual Sentiment Analysis with Interpretability

Authors: Tomoki Ito, Kota Tsubouchi, Hiroki Sakaji, Tatsuo Yamashita, Kiyoshi Izumi4231-4238

AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Using real textual datasets, we experimentally demonstrate that the proposed LEXIL is effective for improving the interpretability of SINN and that the SINN features both the high WCSA ability and high interpretability.3 Experimental Intepretability Evaluation This section experimentally evaluates the proposed method in terms of the interpretability in A) WOSL, B) LWCL, and C) GWCL using real textual datasets.
Researcher Affiliation Collaboration Tomoki Ito,1 Kota Tsubouchi,2 Hiroki Sakaji,1 Tatsuo Yamashita,2 Kiyoshi Izumi1 1Graduate School of Engineering, The University of Tokyo, 2Yahoo Japan Corporation
Pseudocode Yes Algorithm 1 LEXIL: Lexical Initialization Learning
Open Source Code Yes The dataset, code, and details will be available in http://bit.ly/ SINN20190904.
Open Datasets Yes We used the following four textual corpora including reviews and their sentiment tags for evaluation. 1) Eco Rev I and II. These two datasets are composed of comments on current (I) and future (II) economic trends and their positive or negative sentiment tags1. 2) Yahoo reviews. This dataset is composed of comments on stocks and their long (positive) or short (negative) attitude tags, extracted from financial micro-blogs.2 3) Sentiment 140. This dataset contains tweets and their positive or negative sentiment tags.3
Dataset Splits Yes We divided each dataset into training, validation, and test datasets, as outlined in Table 1.Table 1: Dataset details for Text Corpus and Annotated data
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, processor types) used for running experiments.
Software Dependencies No The paper mentions software components like 'LSTM' and 'skip-gram method' but does not provide specific version numbers for any libraries, frameworks, or environments.
Experiment Setup Yes We set the dimension of the hidden and embedding vectors to 200 and epoch to 50 with early stopping. We used the mean score of the five trials for evaluation.