![]() Particularly noteworthy are the problems mentioned by unsatisfied consumers as they can provide a valuable source of knowledge for product developers and thus make the defects of the product eliminated or at least minimized by improving failing functionalities. Identifying the most popular expressions at the exploratory analysis stage made it possible to obtain the most frequently raised issues by satisfied as well as unsatisfied users. According to the results, negative reviews were more often objective than the positive ones as the latter were in majority difficult to categorize. Furthermore, I create my own list of context specific words to better fit the particular case I’m working on.Īnalyzing the negative reviews, over a third of them are objective, and again, only one in eight statements present a subjective view. One additional but crucial step that will help generating better results is using a list of stopwords, which simply means that I remove all the words from my database that don’t bring any meaningful value. The next move is to remove punctuation, emojis and digits as I want to keep alphabetic characters only. It is important in order the machine not to treat two or more identical for human words separately just because they were written differently (e.g. I exclude from the database reviews with the rating 3 given by user as they present neither a positive nor a negative sentiment.Īs a first step in cleaning the data and preparing for the analysis I change all text to lower case. Text pre-processingĪt first, I assume that ratings 1–2 present negative feelings regarding the product while ratings 4–5 a positive one. I assume those are simply human errors thus I decide to exclude them from my database to get more accurate results while applying text classification models later on. To get a better understanding I look into some of them. Analogically, there are reviews with very low rating given by user (1) with a positive sentiment according to the tool used. ![]() ![]() There are some opinions with very high rating given by user (5) but estimated as negative by TextBlob. The question that automatically appears in my head is: Is the review rating received by applying sentiment analysis going to be the same as the one actually given by buyers? After executing the code and analyzing the results I notice the following discrepancies. However, there is a Python tool for sentiment analysis provided by TextBlob which can actually estimate the same using only the content of the comment itself. It means, we can say if the opinion is positive or negative just by looking at its rating. What are the functionalities that disappointed them mostly?Īre the customers expressing more emotional or logical approach in the product evaluation? Sentiment AnalysisĮach review in the database has a corresponding rating 1–5 which helps to quickly understand whether a customer would recommend a product or not. What do customers value in a particular type of product? The code, helped me answer a couple of crucial questions: It’s a great example of real unstructured textual data and thus it’s a perfect set for applying text mining tools and discovering hidden information about the consumers’ preferences and feelings. The comments are concerning Bluetooth earphones bought via the platform. “If you want to understand people, especially your customers… then you have to be able to possess a strong capability to analyze text” - Paul Hoffmannįor my first text mining project with Python I decided to use a database from Amazon (source: ) with over ten thousand customer reviews. How? Shortly, it’s all about text mining. Except for being a valuable source of information for other potential buyers they can become an important component of development for the business. ![]() Have you ever wondered what makes a customer buy a specific product? And later on, what makes them become loyal or abandon the company forever? In the times of rapid growth of e-commerce it seems unbelievably easy to gather huge amounts of feedback in form of online customer reviews. ![]()
0 Comments
Leave a Reply. |