Data4Fashion: Proposing A Sentiment Analysis Tool To Quantify Fashion Image Performance on Instagram
A research summary series.
This is the first of a research summary series which was initiated to bring to the limelight the academic studies done on data science in the context of the fashion industry.
Episode One is based on “A Sentiment Analysis Tool for Determining the Promotional Success of Fashion Images on Instagram”.
Follow and subscribe to get article updates.
It is no news that fashion consumers are spending more time on the internet and that millions of big data are being generated daily. In the same vein, fashion brands are working endlessly to create engaging content in the form of videos, blogs, and images on different social media platforms.
According to Statista (2022), 31% of the global audience on Instagram are between ages 25–34, and over two-thirds of Instagram users are aged 34 and younger, hence making it a great platform for marketers.
However, it is not enough for fashion brands to create content for their users, what would be of more value is a way to analyze the performance of this content and its impact on the buying decision of fashion consumers.
On this premise, I have written a summary of a research paper titled- A Sentiment Analysis Tool for Determining the Promotional Success of Fashion Images on Instagram.
The authors of this paper proposed a sentiment analysis tool to examine the performance of images posted on Instagram accounts of fashion brands in order to determine their impact on fashion consumers.
Interested in knowing what they did?
Keep reading...
Introduction
This paper aims to analyze the impact of images shared on Instagram accounts of fashion houses on the decision of a customer to make a purchase. It examines features of 20 images with the highest number of likes of the top 50 Fashion houses on Instagram in order to classify whether the opinion on them was positive or negative, and assign social values ranging from -1 to +1. The main question asked in this study is if certain visual aesthetics were more fascinating hence the reason for their success in successful media. It is the first of a number of studies investigating sentiment analysis on Instagram.
Literature Review
Numerous studies were cited in relation to the two broad classifications of the techniques used in opinion mining; machine learning methods and lexicon-based approach. The machine learning method can further be categorized as supervised, for example, the Support Vector Machines and Naïve Bayes(Miwari et al. 2015 ), or unsupervised learning, for example, Probabilistic Latent Semantic Analysis (PLSA) (Arora et al. 2015). The third classification of machine learning is semi-supervised learning, which is a relatively new concept that combines the technique of both supervised and unsupervised machine learning (Arora et al. 2015).
On the hand, the Lexicon-based method, which is the approach of interest in this paper, can either be Dictionary-based methods (Feldman 2013) or Corpus-based methods (Arora et al. 2015). Paltoglou & Thelwall (2012) proposed a lexicon-based classifier that predicts the intensity of emotion within a text as positive (+1), negative (-1), or neutral (0) in order to provide predictions that would address issues pertaining to sentiment analysis.
Similarly, Hridoy et al. (2015) suggested a different method that allows for the “utilization and interpretation of Twitter data to determine public opinion proposed another more specific technique”. In this study, Twitter API was used to get the data and data cleaning was performed with Java, and the Stanford Natural Language Processing(SNLP) tool was then used to label the provided data since SNLP provides the grammatical relations between words, and SentiWordNet was used to assign scores for the provided tweets (Hridoy et al. 2015).
On studying different works of literature, Mohamed AbdelFattah et al. (2016), based on evidence, concluded that lexicon-based approaches are better suited when unsupervised learning is in order. Hence the reason for adopting this technique in their research.
Proposed Methodology
The authors of this paper proposed a data collection method that would require the extraction of images uploaded by the selected fashion brands, followed by the determination of the social value of the images using the application of Sentiment Analysis (SA). Fig. 1 shows the proposed workflow.
Two modules and an out phase were proposed in this study, with each having subset steps. The first module is the Data Module.
Data Module
This will involve the extraction of data and pre-processing of the data. Images, comments, likes, and names of brands will be extracted, saved in the MySQL database, and then preprocessed using SNLP.
This process aligns with the approach used by Hridoy et al. (2015) in their research on Twitter texts anaysis.
Sentiment Analysis Module
Following the pre-processing phase, the resulting data will pass through a comment analyzer, image evaluator, and brand evaluator components. The result of this process will be a social value for each comment, ranging from -1 to 1, with 1 being the most positive, -1 as the most negative, and 0 as neutral. Based on the social value score, the images will be evaluated and also be given a score. In this same vein, an overall score for each brand will be derived at the end of the Sentiment Analysis module.
Output
The result of this proposed methodology will be a final social score for individual fashion brands. This value can then be used by to analyze the contents on the Instagram account and decide what image to leave, modify, promote or re-create.
Conclusion
The authors of this paper concluded that by assigning a social value to fashion images, these images can be quantified thus aiding fashion brands to better analyze the performance of their content, and their follower’s reactions and subsequently aid in making informed brand and marketing decisions.
Finally, Mohamed AbdelFattah et al.(2016) stated that the product of this study will be presented to domain experts and the feedback received will inform the necessary adjustments to the study.
Click HERE
to access the full article and Summary pdf
Hope you enjoyed reading the article as much as I enjoyed writing it. Don’t forget to drop your feedback in the comment session and give this article 50 claps!
Connect with me on LinkedIn.
CHEERS!
