Group11 proposal
Contents
Background & Motivation
Twitter, an online social media platform with over 300 million monthly active users, serves as an important customer support platform for businesses. Over 27% of customers already take to Twitter to share product reviews, air their frustrations and connect with brands (ConverSocial, 2016). Research shows that when a customer Tweets at a business and receives a response, they’re willing to spend 3–20% more on an average-priced item from that business in the future (Alton, 2017). Thus, analysing these tweets help to identify best practices and common issues for businesses which are valuable in improving customer experience. However, it is difficult to analyse these unstructured data. Organisations often turn to commercial off the shelf tools to analyse text data and extract important information. Nevertheless, commercial off the shelf tools are costly and unable to customise to the needs of the organisations.
Project Objective
Using airline tweets as a case study, this project aims to use the available text mining packages in R to build an interface for users to perform text analytics(i.e namely topic modeling and sentiment analysis) without the need to code. Users can also visualise the results in an interactive manner to uncover insights in the airline tweets.
Data Source
We have chosen the Customer Support on Twitter dataset from Kaggle. (https://www.kaggle.com/thoughtvector/customer-support-on-twitter)
This dataset comprises of over 2.8 million tweets from 108 companies and over 700,000 unique users from May 2008 to December 2017. As we are focusing on the airline industry as a case study, we will filter for tweets that are related to airlines.
600px
The chart above shows the distribution of outbound tweets by author_id. Some of the airlines we will be looking are circled in red.
The dataset contains both in bound and outbound tweets. However, we will only being using inbound tweets for our analysis as companies would be more interested to find out about customers' feedback and their overall sentiment.
Methodology
For text analysis on the airline tweets, we would first perform text pre-processing namely tokenisation, stop words removal and stemming. The processed tweets will then be analysed via two key techniques: Topic Modelling and Sentiment Analysis.
Topic modelling is about finding a topic in a set of words that are frequently co-occurring together. In this project, we would be looking at Latent Dirichlet allocation (LDA) to derive the topics for the tweets.
For sentiment analysis, we would be focusing more on sentiment polarity classification where airlines would be able to find out if the customers’ have a positive or negative sentiment about the company in general or about their services. Sentiment polarity of the customers’ tweets would be determined via the lexicon approach where we would use any existing pre-compiled lexicons in R or via the classification approach where we will train a binary classifier to predict the sentiment polarity of a new tweet. Available sentiment analysis packages in R such as “sentiment r” would be explored.
Visualisation Features
| Sketch | Description |
|---|---|
| This is the first tab (i.e intro tab) of our R Shiny App that will be shown to our users. In this tab, we will provide a short description on the App to allow users to have a brief idea on what kind of analysis can be done on this App. | |
| This where the user will upload the data to our App per the data requirement. Data requirement will be specified on this tab. Having an upload data feature allows users to analyse different kind of tweets as long as the data structure adhere to our App's data requirement. | |
| The third tab will be about topic modelling results. In this tab, there will be a total of two sub-tabs. The first sub-tab will reflect the model comparison results between LDA and LSA based on the airline and time period of the user’s choice. Thereafter, user will choose between using the LDA or LSA models to derive the topics. Users would also be shown the optimal number of topics so as to give them an idea on the number of topics they should put as input for the next sub-tab which will then generate a list of key words for each topic. | |
| We will use R libraries such as “sentiment r” to analyze the general sentiment of each airline. Before deploying which model to use, we will use a confusion matrix and classification table to evaluate the different models. User are then able compare the general sentiments of different airlines for the chosen time period. |
References
Alton, L. (2017, December 5). 4 tips for providing effective customer support on Twitter. Retrieved from https://business.twitter.com/en/blog/4-tips-for-providing-effective-customer-support-on-Twitter.html
ConverSocial. (2016). The State of Social Customer Service. Retrieved from http://www.conversocial.com/hubfs/Conversocial-Report-The-State-of-Social-Customer-Service-16.pdf
