Natural Language Processing is of particular interest in the fields of commerce and marketing, especially for the analysis of the opinions and sentiments expressed by consumers.
Voice of the Customer
Imagine you have hundreds of thousands or even millions of consumer feedback. All this represents a gold mine of information to improve your business. Voice of the customer analysis helps you understand consumer expectations and transform those expectations into more relevant products and services.
Most of these comments evoke familiar things: for example, a product is too expensive or not organic enough; or there is too much time spent at the checkout. Measuring these signals allows to quantify their evolution over time as they are resolved. Once these signals are removed, there remain plenty of weak signals. Some of them contain nuggets.
Let’s give a few concrete examples identified at one of the largest French distributors. For example, the company increased the turnover of the paper department by 8% in a very simple way. By leaving the diaries at the front of the gondola an extra week in September.
Indeed, many customers wrote to complain that they could no longer find the diary they had come to the store to get for their children at the beginning of the school year.
It can also be used to improve product quality. One type of peach yoghurt had a systematic problem of rotting before the use-by date. The problem was being brought up by only a few people from time to time. This made it imperceptible to the human eye. But the machine detected it, which eventually fixed the problem in the factory.
The exploitation of weak signals is also interesting to anticipate health or legal risks. For example, on a given week, five people complain of bitterness in the mouth and loss of taste after eating pine-nuts. These pine nuts were not edible for humans. Immediate identification of this risk allows the offending product to be removed from the shelves as soon as possible.
Finally, weak signals also help to identify societal trends. For example, the increase of consumer demand for organic halal products was clearly visible as soon as the solution was implemented.
Moderation of Internet user’ reviews
Another interesting use case is the automatic moderation of users’ reviews. A major retailer in the sports sector thus offers its customers the opportunity to leave product feedback on its website. But these comments must respect certain rules: do not quote personal data or competing brands; do not make offensive or inappropriate comments…
Semantic analysis allows to largely automate the moderation process by verifying that these rules are respected. To give an order of magnitude, the system thus automatically validates 70% of the opinions; it also automatically blocks the 10% of opinions that violate one of the rules that must be respected. This allows human teams to focus on the 20% of reviews that require an expert eye to validate or not.
Another example of value creation is the predictive analysis made possible by the analysis of texts from websites. A few years ago, we analyzed the sites of influential bloggers in the world of footwear and fashion. During this study, we detected hundreds of types of shoes, including peep-toes (i.e. shoes with high heels and open rounded toes). We noticed an increasing trend in these mentions over time for this model. We predicted that this type of shoe would sell more the following summer, and it actually happened.
Identifying Social Web Influencers
Another interest of this study was to compute the relative importance of influential bloggers. Just as researchers quote each other, bloggers compare themselves to each other. After collecting 20,000 blog posts in this universe, we computed what is called a centrality graph: the more you are in the center of the graph, the more legitimate you are in the eyes of your peers.
In this case, the blogger symbolized by the red dot in the center of the graph was chosen by a major shoe brand to represent it as a Web 2.0 ambassador.
As we can see in the graph below, this is a powerful approach that also allows a brand to establish a sub-segmentation.
Email response assistants
An example of automation enabled by naturel language processing is the help for answering emails. In retail banking, the system is able to detect questions asked in an email received and then search a knowledge base for the best answers. A draft answer is thus automatically initiated and only has to be validated before sending.
The system can also query the information system to get answers to specific questions; for example, the amount in a bank account. This type of solution allows an account manager to save up to one hour per day.
Chatbot is another example of easy to implement automation, operating 24 hours a day, 7 days a week. It helps to reduce the burden on customer service on repetitive questions by answering within a certain limit to the questions that Internet users ask themselves.
One point that seems essential to us in these two examples is to always be transparent about the automation by making it clear that it is with a machine that we are dialoguing and by possibly proposing to continue the dialogue with a human.
Understanding human language
Most words have several meanings. The presence of several words in the sentence (and in neighboring sentences) creates a context that helps to identify the meaning of a word among those that are possible. Disambiguation algorithms can then exploit the different clues present. A good understanding of the linguistic difficulties that may be encountered is also necessary, as these lexical ambiguities are of different natures.
As you can see, Natural Language Processing uses are very numerous, and it offers a wealth of opportunities to increase productivity, reliability, and make better decisions. We could still demonstrate so many uses!
Read more about about NLP applications: