This is akin to just scratching the surface and missing out on those high-value insights waiting to be discovered. So what should a brand do to capture that low hanging fruit?is contextual mining of text, which identifies and extracts subjective information in the source material and helping a business to understand the social sentiment of their brand, product, or service while monitoring online conversations. However, analysis of social media streams is usually restricted to just basic sentiment analysis and count-based metrics.
With the recent advances in deep learning, the ability of algorithms to analyze text has improved considerably. Creative use of advanced artificial intelligence techniques can be an effective tool for doing in-depth research. We believe it is important to classify incoming customer conversation about a brand based on the following lines:
- Key aspects of a brand’s product and service that customers care about.
- Users’ underlying intentions and reactions concerning those aspects.
These basic concepts become an essential tool for analyzing millions of brand conversations with human-level accuracy when used in combination. In the post, we take the example of Uber and demonstrate how this works. Read On!
Sentiment Analysis is the most common text classification tool that analyses an incoming message and tells whether the underlying sentiment is positive, negative, or neutral. You can input a sentence of your choice and gauge the underlying sentiment by playing with the demo .
Intent analysis steps up the game by analyzing the user’s intention behind a message and identifying whether it relates to an opinion, news, marketing, complaint, suggestion, appreciation, or query.
Analyzing intent of textual data
Contextual Semantic Search(CSS)
Now, this is where things get really interesting. To derive actionable insights, it is important to understand what aspect of the brand a user is discussing. For example, Amazon wants to segregate messages related to late deliveries, billing issues, promotion-related queries, product reviews, etc. On the other hand, Starbucks wants to classify messages based on whether they relate to staff behavior, new coffee flavors, hygiene feedback, online orders, store name, location, etc. But how can one do that?
We introduce an intelligent smart search algorithm called Contextual Semantic Search (a.k.a. CSS). CSS works because it takes thousands of messages and a concept (like Price) as input and filters all the messages that closely match the given concept. The graphic shown below demonstrates how CSS represents a major improvement over existing methods used by the industry.
Existing approach vs. Contextual Semantic Search
A conventional approach for filtering all Price related messages is to do a keyword search on Price and other closely related words like (pricing, charge, $, paid). However, this method is not very effective as it is almost impossible to think of all the relevant keywords and their variants that represent a particular concept. On the other hand, CSS takes the concept (Price) as input and filters all the contextually similar even where the obvious variants of the concept keyword are not mentioned.
For curious people, we would like to give a glimpse of how this works. An AI technique is used to convert every word into a specific point in the hyperspace. The distance between these points is used to identify messages where the context is similar to the concept we are exploring. A visualization of how this looks under the hood can be seen below:
Time to see CSS in action and how it works on comments related to Uber in the examples below:
Similarly, have a look at this tweet:
In both the cases above, the algorithm classifies these messages as contextually related to the concept called Price even though the word Price is not mentioned in these messages.
Uber, the highest valued start-up in the world, has been a pioneer in the sharing economy. Being operational in more than 500 cities worldwide and serving a gigantic user base, Uber gets many feedback, suggestions, and user complaints. Often, social media is the most preferred medium to register such issues. The huge amount of incoming data makes analyzing, categorizing, and generating insights a challenging undertaking.
We analyzed the online conversations on digital media about a few product themes: Cancel, Payment, Price, Safety, and Service.
We took data from the latest comments on Uber’s official Facebook page, Tweets mentioning Uber, and the latest news articles around Uber for a wide coverage of data sources. Here’s a distribution of data points across all the channels:
- Facebook: 34,173 Comments
- Twitter: 21,603 Tweets
- News: 4,245 Articles
We ran the Contextual Semantic Search algorithm on the same dataset, taking the aforementioned categories into account (Cancel, Payment, Price, Safety, and Service).
Breakdown of Sentiment for Categories
Noticeably, comments related to all the categories have negative sentiment, majorly, bar one. The number of positive comments related to Price has outnumbered the negative ones. To dig deeper, we analyzed the intent of these comments. Facebook being a social platform, the comments are crowded random content, news shares, marketing promotional content, and spam/junk/unrelated content. Have a look at the intent analysis on the Facebook comments:
- Intent analysis of Facebook comments
- Intent analysis of Facebook comments
Thus, we removed all such irrelevant intent categories and reproduced the result:
Filtered Sentiment Analysis
There is a noticeable change in the sentiment attached to each category, especially in Price related comments, where the number of positive comments has dropped from 46% to 29%.
This gives us a glimpse of how CSS can generate in-depth insights from digital media. A brand can thus analyze such Tweets and build upon the positive points from them or get feedback from the negative ones.
A similar analysis was done for crawled Tweets. In the initial analysis, Payment and Safety related Tweets had a mixed sentiment.
Category wise sentiment analysis
To understand real user opinions, complaints and suggestions, we have to filter the unrelated Tweets(Spam, junk, marketing, news, and random) again:
There is a remarkable reduction in the number of positive Payment related Tweets. Also, there is a significant drop in the number of positive Tweets for the category Safety(and related keywords.)
Additionally, Cancel, Payment, and Service (and related words) are the most discussed topics in Twitter comments. It seems that people talked most about drivers canceling their rides and the cancellation fee charged to them. Have a look at this Tweet:
A brand like Uber can rely on such insights and act upon the most critical topics. For example, Service-related Tweets carried the lowest percentage of positive Tweets and the highest percentage of Negative ones. Uber can thus analyze such Tweets and act upon them to improve service quality.
Sentiment Analysis for News headlines
Understandably so, Safety has been the most talked about topic in the news. Interestingly, news sentiment is positive overall and individually in each category as well.
We classified news based on their popularity score as well. The popularity score is attributed to the share count of the article on different social media channels. Here’s a list of top news articles:
The age of getting meaningful insights from social media data has now arrived with the advance in technology. The Uber case study gives you a glimpse of the power of Contextual Semantic Search. It’s time for your organization to move beyond overall sentiment and count-based metrics. Companies have been leveraging data power lately, but to get the deepest of the information, you have to leverage AI, Deep learning, and intelligent classifiers like Contextual Semantic Search and . At Karna, you can contact us to license our technology or get a customized dashboard for generating meaningful insights from digital media. You can check the demo .
We hope you liked the article. Pleasefor a free ParallelDots account to start your AI journey. You can also check demos of ParallelDots APIs .