Source-linked AI summary
Fuzzy Based Implicit Sentiment Analysis on Quantitative Sentences
Amir Hossein Yazdavar, Monireh Ebrahimi, Naomie Salim
TL;DR
Objective statements with quantitative medical values can imply sentiment, yet research has focused largely on subjective sentences containing opinion keywords. The paper uses a fuzzy rule-based expert system to classify such statements in drug reviews; its findings describe the approach as promising while acknowledging polarity errors.
Problem
Research has focused largely on subjective sentences with opinion keywords, while sentiment orientation in quantitative drug-review statements remains underexplored.
Source: §1 · p. 1, §3 · pp. 3–4Method
The system combines knowledge acquisition and a fuzzy rule-based expert system to classify quantitative medical sentences as positive, negative, or neutral toward drug effectiveness.
Source: §4.5 · p. 7, §4.5.2 · p. 7Results
The authors report that the fuzzy system is promising for medical sentiment analysis but cannot correctly determine the polarity of some statements.
Source: §6 · p. 9Takeaways & Limitations
The authors conclude that rule-based fuzzy methods can distinguish some sentiment-bearing objective sentences from non-opinionated ones in drug reviews.
Source: §7 · p. 11The evaluation discussion identifies temporal and conditional statements, factual misclassifications, coreference, and verb-less sentences as challenges for the system and extraction algorithm.
Source: §5 · pp. 8–9
Abstract
from arXiv · showhide
With the rapid growth of social media on the web, emotional polarity computation has become a flourishing frontier in the text mining community. However, it is challenging to understand the latest trends and summarize the state or general opinions about products due to the big diversity and size of social media data and this creates the need of automated and real time opinion extraction and mining. On the other hand, the bulk of current research has been devoted to study the subjective sentences which contain opinion keywords and limited work has been reported for objective statements that imply sentiment. In this paper, fuzzy based knowledge engineering model has been developed for sentiment classification of special group of such sentences including the change or deviation from desired range or value. Drug reviews are the rich source of such statements. Therefore, in this research, some experiments were carried out on patient's reviews on several different cholesterol lowering drugs to determine their sentiment polarity. The main conclusion through this study is, in order to increase the accuracy level of existing drug opinion mining systems, objective sentences which imply opinion should be taken into account. Our experimental results demonstrate that our proposed model obtains over 72 percent F1 value.
1. Introduction
The introduction identifies implicit sentiment in objective quantitative statements as a gap in opinion mining and motivates fuzzy methods for handling medical values and their changes. The paper applies this approach to classify such statements as opinionated or non-opinionated and determine their polarity.
1. Introduction
- Implicit opinions are conveyed by objective statements rather than opinion keywords, a category that has received limited research despite its domain- and context-dependence.The introduction contrasts these statements with explicit opinions in subjective sentences containing opinion keywords.Source: §1 · p. 1
- A value change can signal improvement or worsening through its direction and destination, but changes may also be factual and non-opinionated.For example, a cholesterol decrease into a desirable range is positive, while a dosage change can simply report a fact.Source: §1 · pp. 1–2
- Medical measurements have imprecise ranges, and changes can indicate improvement, stability, or exacerbation even when the new value remains outside the normal range.The authors motivate fuzzy logic to represent value ranges and changes as slight, medium, or high.Source: §1 · p. 2
- The paper uses fuzzy set theory to classify quantitative objective sentences as opinionated or non-opinionated and determine whether their sentiment is positive or negative.The authors describe this as the study's main contribution.Source: §1 · p. 2
- The study frames four tasks: extracting numeric variables, linking numbers to entities, identifying sentiment-bearing sentences, and determining polarity.It situates these tasks in the broader need for automated opinion extraction from large volumes of online reviews.Source: §1 · pp. 1–2
2. Related Works
Related work covers general sentiment methods, drug-review challenges, biomedical entity extraction, and fuzzy approaches to medical uncertainty. The section identifies sentiment orientation in number-containing drug reviews as an unresolved problem addressed by this paper.
2. Related Works
- General sentiment methods include pattern-and-PMI classification (Turney, 2002), log-likelihood with adjective seeds, and synonym/antonym bootstrapping.Aspect sentiment research also includes supervised and lexicon-based approaches, with training-data dependence noted for supervised methods.Source: §2 · p. 2
- Drug-review research treats side effects as implicit negative evidence and notes that when a fact occurs relative to drug use can affect whether it expresses opinion.Other cited work combines bio-entity relationship features with lexicon-based polarity prediction.Source: §2 · p. 3
- Prior biomedical work extracts numeric fields with regular expressions (Bhatia, et al., 2010) and uses GATE rules for domain-based syntactic and semantic co-occurrence.Entity-linking approaches include co-occurrence, linguistic rules, and dependency or link grammar parsing; this paper adopts linguistic patterns and heuristics.Source: §2 · p. 3
- Fuzzy systems address medical uncertainty because measurements such as blood pressure can lack a single precise threshold and may require flexible intervals.The section cites a survey of fuzzy applications in medicine (Abbodet al., 2001).Source: §2 · p. 3
- Sentiment orientation in drug-review sentences containing numbers remained unexplored, motivating the paper's fuzzy-method approach.The section presents this as a major unaddressed issue.Source: §2 · p. 3
3. Problem definition
The problem is to distinguish sentiment-bearing quantitative statements in drug reviews from factual numeric statements, then assign polarity to the former. The paper focuses on biomedical measurements whose values can express health status.
3. Problem definition
- Objective quantitative statements can express sentiment through two values for one medical term, an explicitly stated change, or one value paired with a change verb.Examples include LDL changing from 175 to 105, cholesterol falling by 25 points, or cholesterol falling to 160.Source: §3 · p. 3
- Numeric statements are not necessarily opinions: dosages, drug-use counts, and disease duration may state facts and need filtering.A requested cholesterol target can remain factual despite a change verb; weight values also require context to distinguish an opinion from a measurement report.Source: §3 · pp. 3–4
- The target entities are biomedical measurements such as blood pressure, weight, fasting glucose, total cholesterol, LDL, HDL, and triglycerides.Their numeric values are used to monitor health and can implicitly convey sentiment in biomedical text.Source: §3 · p. 4
4. Proposed Approach
The approach collects cholesterol-drug reviews, extracts and maps quantitative medical terms and values, then uses expert-informed fuzzy rules to classify sentence sentiment. The system’s inputs include drug type, dosage, and duration, while its output is positive, neutral, or negative sentiment.
4.1 Overview
- The workflow builds a patient-review corpus with an expert-created gold standard, extracts medical terms and values, maps them, and evaluates polarity using precision and recall.The corpus includes numerated opinionated sentences, with polarity distributions summarized in Tables 2 and 3.Source: §4.1 · p. 4, §4.2 · p. 4
4.2 Collecting Drug Reviews
- The study collected 210 reviews from Ask a Patient covering five statin categories and niacin.Medical experts helped account for differences in drug characteristics and patient responses.Source: §4.2 · p. 4
4.3 Preprocessing
- GATE and JAPE rules extract quantitative medical terms, values, change verbs, drug names, and dosages, then associate terms with values.Regular-expression rules are designed to handle multiple term-value pairs and verbless entries such as “HDL: 175.”Source: §4.3 · p. 5
4.4 Fuzzy Logic Approach
- Fuzzy set theory maps sentence inputs through expert-generated rules to determine sentiment orientation in drug reviews.The approach specifies input and output variables, their fuzzy sets and membership functions, and an inference process with defuzzification.Source: §4.4 · p. 5
4.5 Development of Fuzzy Logic Technique
- Expert-informed fuzzy rules classify sentence polarity as positive, neutral, or negative after medical terms and their values are extracted.The system uses a knowledge base, inference engine, and fuzzified inputs; its fuzzy rules were verified by another doctor and conflicts resolved by expert consensus.Source: §4.5 · p. 7, §4.5.2 · pp. 7–8
4.5.1 Knowledge Acquisition by Interview and Literature
- Medical experts considered drug type, dosage, and duration when judging drug effectiveness and designing the expert system.Experts identified these factors through interviews, with drug type and dosage also informed by literature on cholesterol-lowering drugs.Source: §4.5.1 · p. 7
4.5.2 Structure of Proposed Fuzzy Expert System
- The system uses trapezoid membership functions for fuzzy variables, with medical ranges informed by literature and doctors’ opinions.Sentiment is represented by negative, neutral, and positive fuzzy sets; the system uses Mamdani inference and centroid defuzzification.Source: §4.5.2 · pp. 7–8
- A reported error case attributes a cholesterol decrease to exercise rather than the drug, motivating subject extraction to address such cases.The authors examine false positives and false negatives using a confusion matrix.Source: §4.5.2 · p. 8
5. Results
The evaluation uses a human-expert gold standard to assess extraction and sentiment classification, while the reported discussion identifies temporal cues and difficult sentence structures as challenges.
5. Results
- Human experts labeled corpus sentences to build a gold standard for evaluating value and medical-term extraction, term–value pairing, and sentiment classification.Precision, recall, and F-measure were used; overall precision and recall are reported in Table 7.Source: §5 · p. 8
- Temporal context can affect polarity: a slight HDL increase over a short period is treated as positive, while some factual statements are classified as opinionated.Conditional sentences are also identified as requiring special treatment.Source: §5 · p. 8
- Coreference and verb-less sentences challenge extraction, and some cases are not handled comprehensively enough to extract entities properly.Source: §5 · p. 9
- Dense sequences of medical measurements require special patterns to map each medical term to its corresponding value; the extraction method does not handle all such cases.Source: §5 · p. 9
6. Discussion
The discussion presents fuzzy knowledge engineering as promising for medical sentiment analysis, but notes classification errors and dependencies on rule construction and extraction.
6. Discussion
- The discussion identifies expert-annotated corpus development and sentiment polarity from variable changes as contributions, describing the latter as unexplored in prior literature.Source: §6 · p. 9
- The fuzzy system is described as promising for medical vagueness, but some statements receive incorrect polarity and broader expert consensus is needed to expand its rules.The paper also suggests combining fuzzy knowledge-based methods with machine learning; rule generation was time-consuming.Source: §6 · p. 9
- Conditional sentences are among the complex cases that the polarity classifier assigns to the wrong category, motivating expansion of the algorithm.Source: §6 · p. 9
- System performance depends on extraction that misses cases involving verb-less text, grammatical mistakes, and typographical errors; supervised or semi-supervised extraction is proposed as an alternative.Source: §6 · p. 9
- Ontology-based extraction with deep parsing is proposed to improve extraction reliability by accounting for intricate relationships among entities.Source: §6 · pp. 9–10
7. Conclusion
The conclusion presents a rule-based fuzzy technique for identifying sentiment in drug reviews from numeric-attribute changes and reports corpus and annotation needs for improving performance.
7. Conclusion
- The technique uses rule-based fuzzy set theory to distinguish opinionated objective sentences from non-opinionated ones and determine their sentiment in drug reviews.The authors say it helps explain reviewers’ attitudes toward target-drug efficacy and could serve as a subtask in a drug-efficiency meter using customer feedback.Source: §7 · p. 11
- The authors describe the results as encouraging, while stating that a larger corpus of quantitative medical terms and more expert annotators could improve performance.Source: §7 · p. 11
- More medical knowledge encoded in fuzzy rules and hybrid approaches are also proposed as ways to improve results.Source: §7 · p. 11