Sentiment Sensitive Machines for Quicker Action

How many times have we been confused with the plethora of reviews and comments on items, while shopping online? These days, seldom do we reach out to our trusted acquaintances for opinions before buying an item, unless probably it is a big ticket one. Instead, these online opinions from unknown people are what shape our buying decisions. Whenever we browse for a prospective purchase, we are (almost always) bombarded with millions of reviews on the product page. Have you ever wondered what retailing platforms do with these reviews? The purpose is not to just generate content. There is a larger picture associated with it.

Comments and reviews play a significant role for retailers to understand consumer sentiments. Natural language processing (NLP), data mining and text mining capabilities are the three pillars of sentiment analysis. These three aspects together derive meaningful information from human language. Sentiment analysis can categorize a review as positive, negative or neutral. It can also consider the meaning of words with reference to a specific context. Retailing businesses can now understand what their customers intend to convey and their reaction towards a product in real time. Certain sentiment analysis engines rely on individual keyword analysis or more manual methods of identifying sentiment, while there are techniques which analyze the full context of a post or comments, including basic forms of sarcasm and irony. This gives a significant edge when it comes to real-time sentiment analysis.

IBM, Hewlett Packard and Accenture to name a few are assignees of interesting patents in this domain. IBM’s patent US9,418,375B1 discloses a method for determining rating of a product based on sentiments reflected by comments in social media sites, blogs, websites and reviews. In response to a review from a customer, an effective product rating is calculated after depicting positive and negative ratings. Thereafter, one or more products are recommended following adjusted ratings.

Fig. 3 of US9,418,375B1 depicts a graph highlighting difference between initial and adjusted product ratings; the initial ratings are adjusted basis introduction of a similar product.

IBM’s patented technique helps to estimate the scale of a product’s acceptance and allows the retailer to decide on paid advertisement of the product on other forums. Retailers no longer merely focus on product specifications to compare their products vis-à-vis competitors; acting upon users’ opinions has formed an important aspect for advertisement.

US9,595,053B1, another patent by IBM, demonstrates a method for recommending products to customers based on sentiment analysis.

Customers are increasingly taking to social media to express their opinion on products. The expression can extend to the level of stocks in various stores, comparison of customer experience and which makes for a better bet across various aspects of shopping. These are ultimately tagged to customer sentiments. Sentiment analysis helps retailers to achieve granularity in their assessment of reasons fuelling customers dissatisfaction. By using a learned machine, such data can be classified according to its sentiment polarity i.e. positive, negative and neutral. These machines, at times, are brilliant enough to understand sarcastic comments. For example, if a user took to social media to write: “I could not gift my mother on Mother’s Day; thanks to your delayed delivery”. An intelligent algorithm will mark this in the negative and help the retailer to improve services.

“In The Chat Communications Inc.” owns a patent US9,639,902B2 that discloses a method to target customers through social networks. Social media activity of customers and their sentiments towards a company are monitored in real time. A snapshot below shows a method in which live feed from customers is analysed to determine number of customers interested in buying products or services of the company.

Fig 5A of US9,639,902B2 shows a screen that includes a summary of social media activities related to companies and customer.

Though sentiment analysis using NLP techniques is an interesting track to follow, it is limited by size of the training data set and the irrationality of sentiments being displayed by users. Opinions vary from person to person for the same product and the same service. Several factors may also influence sentiments; an argument with a friend might end in a negative tweet about a product, for which there could have been a neutral opinion. Therefore, it is of utmost importance that the systems are trained with a large sample dataset, which gives the opportunity to find an average and not rely on outliers.

Companies have been using consumer feedback to improve their products for decades now. Thanks to the growing usage of social media by consumers to voice their opinions; customer feedback and perception of a brand can now be gauged more easily than before. Companies can use this data to understand consumers’ requirements and improve their strategies accordingly.

Clearly, sentiment analysis aids organisations’ strategies by giving organisations the much-needed insights on their customers.

(Featured image is for representational purpose and has been sourced from https://www.flickr.com/photos/gleonhard/28664657883

Modhura Roy
Modhura Roy

An IP lawyer with expertise in varied technology implementations in the retail domain. Modhura likes to unwind herself by traveling and reading the books of her favorite authors Amitav Ghosh and V. S. Naipaul.


Akhilesh Kumar
Akhilesh Kumar

"Ability is nothing without opportunity" is what this young researcher believes in. Apart from keeping himself updated with latest technologies, Akhilesh likes to devote his time to social causes.


Post a Comment

Your email address will not be published. Required fields are marked *