About
An app with information to show
At Frasal we have WO, a product that aims to make travel planning easier for end-users. It also offers easy ways to share information and travel with other users. But more than anything, it tries to solve common problems for users who decide to go on a trip, such as lack of information, over-information, traveling in groups and several other inconveniences.
But the main complexity of this solution lies in managing millions of points of interest on the map, each with generic information, images, and user comments. This is the fort of WO
Challenge
Show relevant and summarized information to the user
One of WO's differentials over other tools that make travel planning easier is that it tries to display valuable information for the user. Neither more nor less information. In this case, the amount of reviews that exist from a point of interest such as a hotel comes into play. Although a hotel has 4 stars out of 5, as users we are also forced to read many reviews before making sure that those 4 stars are worth it. The intention of WO is to facilitate this step by offering a summary of the existing comments and providing them to the user in a simple and direct way as a real data, just as if he had summarized it himself
Solution
Using Natural Language Processing
A pipeline was created where the information from the reviews would pass. At each stage of the model different information is processed and then used in the next stage. We extract first the entities (such as environment, food, aromas, comfort, internet, etc.), and then the polarity of the comment is extracted (negative, positive, neutral).
Having this information, the model is capable of analyzing a set of reviews and determining what each comment is about and whether the opinion is positive or negative. In this way we can summarize all the comments, saying the good and not so good of each point of interest.
In order to determine the summary of a point of interest, the application first consults an external provider for the comments that users have uploaded for years, and then sends them to the Artificial Intelligence service, which processes them as previously stated.
Results
A successful service to use
The solution can return millions of points from our database quadrants and, clicking to any poi we get in a second the summary from hundreds of reviews
Today the review analysis service is used internally from the Wo application and from the APIs that we have exposed. In addition, the service is being adapted to offer it directly on the internet for those who want to use it with the information they have available, without the need to use our data providers
