For verification of this paper, please visit on http://www.socialresearchfoundation.com/anthology.php#8

Sentiment Analysis is a quantitative tool used to assess the polarity of a document by the way of scaling on a multiway scale, it helps the analyst in finding author's attitude towards the topic. This paper is an attempt to fit in an index that can be used to rank references on a scale of 0 to 10, zero being least useful. We believe that is Sensitivity index is computed for a sufficiently large number of texts it can become instrumental in ranking texts according to their significance with respect to a certain topic. Assigning such ranks will fasten the pace of research and will be able to direct the researcher to the most relevant studies in research.

The objective of this paper is to study of construction of the Sensitivity index.

Beginning with initial works of Turney (2002), Pang (2002) and moving on to more extensive later studies by various other authors and analysts we see an extensive use of techniques evolved in understanding movie reviews. We, however, seek to use the framework of this analysis for benefit of the researcher and research in varied fields. We believe that availability of such a framework would be instrumental in saving time and resources. If we have a framework that catches the sentiment or the intent of any writing it speeds up the research as it helps the researcher in zeroing in upon the most useful references. We propose to fit in an index that can be used to rank references on a scale of 0 to 10, zero being least useful. However, it must be understood that sentiments basically refer to feelings that are held in context, and thus any analysis of sentiments will turn out to be a value judgement. Even though we may assign numeric values to these judgements in the process of fitting an index, our intent remains only to rank them. This approach identifies more with the ordinal approaches as compared to the cardinal approaches.

The first step in construction of the sensitivity index is to find out word sensitivity (S), for this we need to identify the target group of words/word and assign sensitivity value to it on a scale of 1 to 10 (one being the least sensitive or not sensitive). This gives us the Sensitivity scale {1 to 10} in whole numbers. The target word or the group of words is identified to be ‘sensitive’ only if the sensitivity value is greater than 1. Next, we assign a symbol ‘S’ to the sensitive word, where S is a direct function of assigned sensitivity say a.
S = f(lna)
We take natural log of the sensitivity value so that at word sensitivity would be zero and in case a value is 10, sensitivity of the word would be pegged at 1. The value that ‘word sensitivity’ takes thus ranges from 0 to 1. We can similarly find out ‘word sensitivity’ for all the target words, where ranges between 1 to n and n is the total word count of the text. In context, one can also work out the range of word sensitivity for the text by finding out the lowest and highest values assumed by the target words. This range will give us the limits between which the text in question shows sensitivity.
To estimate the text sensitivity (TS), we begin by estimating the sensitive word count (SWC), which is merely a sum total of number of all the sensitive words found in the text.

Given the Sensitive Word Count we can find out the Text Sensitivity Ratio by dividing the sensitive word count (SWC) by the total word count of the text (n).

For an estimate of polarities of mention we will need to modify the sensitivity scale instead of assigning values between {1 to 10} the values will need to range between {-10 to +10} where 0 does not belong to the set. The modified scale will thus be

Such an exercise, however, calls for a normalization of data before we compute the word sensitivity value.
Estimating the Sensitivity Index for a text (SI) can be done by taking a compendium of sensitivity ratios (SR) for different attributes of the text

Here we take simple arithmetic mean across the sensitivity ratio of the text, title, comments received on the text and the references as listed by the author. We are assigning a weight of 2/3 to the sensitivity ratio of the text because it is the text that is being referred to and is most important to the researcher, but other attributes of the writing cannot be ignored, for instance, title shows the intent of the author and should be looked into. Similarly, the comments that are received on the text show how the reader takes to the text and what feelings are evoked in the mind of reader and therefore the sensitivity ratio of the comments must be looked into. Likewise, the sensitivity ratio of the references used by the author are important.
Strengths and Problem Areas of The Sensitivity Index
While constructing the sensitivity index we took care to formulate the index in a manner that it remains additively decomposable. That is, its components can be separately estimated. This gives us the benefit of being able to identify the sensitivity of the title, the text, and the references separately. Our scaling is based on widely used and acknowledged Cantril’s index, that assigns sensitivity on a scale of 1 to 10. This keeps scaling easy and research friendly and assigns a value of zero to word sensitivity when the assigned sensitivity is 1. Further, all the components of the index have face value, moreover the index can be validated using any target text.
The problem areas as we have identified so far are that this index is based on manual assignment of value of sensitivity and will thus attain different values on being run by different researchers. To counter this problem, we must focus on order in which the text is ranked with respect to other texts and not on the actual numeric value attained. As a corollary to the first point mentioned, this index is prone to researcher’s biases. Manual computation also means that such indexing gets complicated in case of larger total word count. Here, we can explore the possibility of using artificial intelligence systems. Further the index may not always give correct results in case of polarities of mention.

This paper introduces the Sensitivity index which if computed for a sufficiently large number of texts can become instrumental in ranking texts according to their significance with respect to a certain topic. Assigning such ranks will fasten the pace of research and will be able to direct the researcher to the most relevant studies in the area of research. With advent of AI, using this index does not look like too distant a possibility. As long as we keep a track of the negatives mentioned and remember that sentiment analysis and the sensitivity index is an offshoot of ‘non computable’ value judgements we should be able to use the index with comfort and for benefit of research.

1. Crossman, Ashley. "How to Construct an Index for Research." ThoughtCo, Aug. 27, 2020, thoughtco.com/index-for-research-3026543. 2. D Sreejith et al, “Sentiment Analysis of English Literature using Rasa-Oriented Semantic Ontology” Indian Journal of Science and Technology, Vol 10(24), DOI: 10.17485/ijst/2017/v10i24/96498, June 2017 3. G. Dharani Devi and Dr. S .Kamalakkannan, “Literature Review on Sentiment Analysis in Social Media: Open Challenges toward Applications” International Journal of Advanced Science and Technology Vol. 29, No. 7, (2020), pp. 1462-1471 4. Handbook on Constructing Composite Indicators, Methodology and User Guide, OECD 2008 5. Louis, Annie, "Predicting Text Quality: Metrics for Content, Organization and Reader Interest" (2013). Publicly Accessible Penn Dissertations. 665. http://repository.upenn.edu/edissertations/665 6. Pang Bo et al, “Thumbs up? Sentiment Classification Using Machine Learning Techniques”, Proceedings of EMNLP 2002, pp. 79-86. 7. Pang, Bo and Lillia Lee, “Opinion mining and sentiment analysis” Foundations and Trends in Information Retrival 2008, 2(1-2):1-135 8. Peter D. Turney, (2002), Thumbs Up or Thumbs Down? Semantic Orientation Applied to Unsupervised Classification of Reviews, Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics (ACL), Philadelphia, pp. 417-424 9. Turney, Peter D. and Michael L. Littman, “Measuring Praise and Criticism: Inference of Semantic Orientation from Association” ACM Transactions on Information Systems 2003 21:314-346 10. M. Taboada, J. Brooke, M. Tofiloski, K. Voll , and M. Stede, (2011), Lexicon-based methods for sentiment analysis, Association for Computational Linguistics, vol. 37 11. Gautam Kumar, Pawan kumar Goel, Sanjeev kumar Chauhan, Anand kumar Pandey, (2012), Opinion mining and summarization for customer reviews, International Journal of Engineering Science and Technology (IJEST), Vol. 4 No.08, ISSN: 0975-5462 12. Welle T et al, “World Risk Index 2012: concept, updating and results”, Works AD (ed) World Risk Report 2012: Environmental Degradation and Disasters, Berlin, pp 11–26 13. Zulfadzli Drus, Haliyana Khalid, “Sentiment Analysis in Social Media and Its Application: Systematic Literature Review” The Fifth Information Systems International Conference 2019. Available online at www.sciencedirect.com