Delta TFIDF: An Improved Feature Space for Sentiment Analysis

By Justin Martineau, Timothy W. Finin. International Conference on Web and Social Media, 2009

392 citations recorded by the source corpus.

Mining opinions and sentiment from social networking sites is a popular application for social media systems. Common approaches use a machine learning system with a bag of words feature set. We present Delta TFIDF, an intuitive general purpose technique to efficiently weight word scores before classification. Delta TFIDF is easy to compute, implement, and understand. We use Support Vector Machines to show that Delta TFIDF significantly improves accuracy for sentiment analysis problems using three well known data sets.

Subfields: Artificial Intelligence

View Paper: https://www.semanticscholar.org/paper/5ef2edfe51d3d768b1d89ad7e74e4ce8e55d1d49

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

Citations in the corpus

Weighted Neural Bag-of-n-grams Model: New Baselines for Text Classification (2016, methodology): “However, in sparse case, heuristic weighting techniques designed by humans are shown to be able to bring signiﬁcant improvements over raw BOW representation (Wang and Manning, 2012; Martineau and Finin, 2009).”
Deep Learning--based Text Classification (2020, neutral): “Δ idf [210] 88.”
Baselines and Bigrams: Simple, Good Sentiment and Topic Classification (2012, neutral): “Valence Shifter : (Kennedy and Inkpen, 2006). tf. ∆ idf : (Martineau and Finin, 2009).”
Learning Word Vectors for Sentiment Analysis (2011, methodology): “In all experiments, we use this weighting to get multi-word representations (Kennedy and Inkpen, 2006) 86.20 N/A N/A tf. ∆ idf Weighting (Martineau and Finin, 2009) 88.10 N/A N/A Appraisal Taxonomy (Whitelaw et al., 2005) 90.20 N/A N/A Table 2: Classiﬁcation accuracy on three tasks.”
A review of natural language processing techniques for opinion mining systems (2017, methodology): “Information retrieve methods are commonly used to construct feature space for machine learning approaches, such as unigram and bigram with tf-idf weights [35, 36, 37].”
Beneath the Tip of the Iceberg: Current Challenges and New Directions in Sentiment Analysis Research (2020, background): “Throughout the development of sentiment analysis, MLbased approaches–both supervised and unsupervised–have employed myriad of algorithms that include SVMs, Moraes et al. [152], Naive Bayes Classifiers, Tan et al. [200], nearest neighbour, Moghaddam & Ester [147], combined with features that range fr”
Large Language Models for Information Retrieval: A Survey (2023, neutral): “The term “document” will henceforth refer to any text-based content subject to retrieve, including both long articles and short passages.”
Multi-Dimensional Sentiment Analysis with Learned Representations (2012, background): “We also note that delta inverse document frequency weighting, which has been shown to sometimes perform well in sentiment (Martineau and Finin, 2009), does not extend easily to multi-dimensional notions of sentiment.”
FinBERT: Financial Sentiment Analysis with Pre-trained Language Models (2019, methodology): “We can divide the recent efforts into two groups: 1) Machine learning methods with features extracted from text with "word counting" [1, 19, 28, 30], 2) Deep learning methods, where text is represented by a sequence of embeddings [2, 25, 32].”
Improvement methods for stock market prediction using financial news articles (2016, methodology): “Several terms weighting methods have been proposed by researchers in the text mining field and we will used delta TF-IDF method [12] instead of the normal TF-IDF, the goal of this new algorithm was to increase the importance of the word which unevenly distributed between positive and negative…”