Papers › An Empirical Evaluation of Word Embedding Models for Subjectivity Analysis Tasks

An Empirical Evaluation of Word Embedding Models for Subjectivity Analysis Tasks

6 Apr 2021IEEE International Conference on Advances in Electrical, Computing, Communication and Sustainable Technologies (ICAECT) 2021 4archive 2025-07-28

Ritika Nandi, Geetha Maiya, Priya Kamath, Shashank Shekhar

It is a clearly established fact that good categorization results are heavily dependent on representation techniques. Text representation is a necessity that must be fulfilled before working on any text analysis task since it creates a baseline which even advanced machine learning models fail to compensate. This paper aims to comprehensively analyze and quantitatively evaluate the various models to represent text in order to perform Subjectivity Analysis. We implement a diverse array of models on the Cornell Subjectivity Dataset. It is worth noting that the BERT Language Model gives much better results than any other model but is significantly computationally expensive than the other approaches. We obtained state-of-the-art results on the subjectivity task by fine-tuning the BERT Language Model. This can open up a lot of new avenues and potentially lead to a specialized model inspired by BERT dedicated to subjectivity analysis.

PaperPDFCode

Code

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

Language ModelingLanguage ModellingSubjectivity Analysis

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Subjectivity Analysis SUBJ BERT-Base + CLR + LSTM Accuracy 97.30 #2 of 19 Archive leaderboard report
Subjectivity Analysis SUBJ BERT-Base + LSTM Accuracy 96.60 #4 of 19 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Methods

AdamAttentionAttention DropoutBERTDense ConnectionsDropoutLayer NormalizationLinear LayerLinear Warmup With Linear DecayMulti-Head AttentionResidual ConnectionSoftmaxWeight DecayWordPiece

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections