Browse Prior Art Database

Sentiment Classification based on Judge's overlap with corpus

IP.com Disclosure Number: IPCOM000245765D
Publication Date: 2016-Apr-06
Document File: 3 page(s) / 69K

Publishing Venue

The IP.com Prior Art Database

Abstract

Sentiment is subjective. It depends on the culture and people involved. This would indicate that there will always be a grey area that the current methodologies of using fine-tuned dictionaries does not address. This has led to the proposed idea of taking into account the expertise levels and trustworthiness of those who judge the sentiment of a body of text (corpus). The accounting of expertise levels will more accurately deal with the grey areas by matching attributes of the corpus against attributes of the work of the judge. It also deals with scoring issues when applying judgements from multiple judges by weighting them based on their expertise level/overlap.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 44% of the total text.

Page 01 of 3

Sentiment Classification based on Judge's overlap with corpus

Executive Summary

The problem to be addressed is increasing sentiment classification accuracy using social data. Accuracy is based on the level of relation (expertise) to the corpus examined and trustworthiness (eminence) of the examiner instead of pre-defined dictionaries.

Abstract

    The approach to Sentiment Analysis, in general, has been to apply a pre-defined objective view. The objective view is usually composed of dictionaries. The dictionaries can be static and pre-populated or dynamic and ever-revising. Sentiment is subjective. As a person, it depends on what your view of the world is. As a company, it depends on defined business goals. When achieving goals, the company will rely on experts. Therefore, as a company, if an expert says that something is positive or negative for the company, then that opinion is more reliable than Vox Populi. Sentiment Analysis should be based on the people we consider relevant experts, not only experts in Sentiment. The problem with current approaches is that domains do not address "grey areas". "Grey areas" are usually introduced when people perform many tasks for the company and their positions evolve.

    This may also apply to categorization in general but the initial focus will be on sentiment as it's a bi-polar categorization problem where Positive and Negative are the poles, Neutral is in the middle if there is little interference and Ambivalent is also in the middle but signifies strong interference. Interference, in this case, is how often the overall score transitions between positive and negative. In the case of categorization, the measurement is based on other pre-defined labels. Based on the overlap, the reliability that the label is correct is the score. This would require the extension that there could be multiple scores or a score combination (modelled as a vector). The accuracy could then be thought of more generally as the acceptability of the answer.

Example Scenario

    A company is understaffed. Person A takes on project management tasks in addition to engineering tasks. The company is bought out and joins a larger organisation. The larger company has no one who had done this. Person B has always been a project manager and Person C has always been an engineer. When a quick survey is conducted. In the case of questions about the interactions between engineers and project managers, Person A should be more reliable. However, if the question is about engineering only then Person C is more reliable than Person A. The same applied to Person B being more reliable than Person A for project management only questions.

Core Idea

    The mechanism is broken up into 2 parts: • Per-Judge • Overall totalling
The Per-Judge mechanism focuses on the overlap in order to determine expertise and the reinforcement of that judgement by peers to demonstrate eminence. The level of overlap is based on attributes of the text which can be taken fr...