Our client's goal is to be close to the aerospace community by providing a technical solution that is useful to them. To do this, they had an online tool that allows the user to find components in a database of more than 15 million records. In addition, they can consult all kinds of technical specifications in a comprehensive, accurate and detailed manner. On the other hand, they also can compare the different options and perform in-depth technical analysis to obtain a complete list of those components of interest.

Challenge

Developing a search system in a database of more than 15 million components would be complicated from a usability point of view. But in this case, it would be even more complicated for two reasons:

On the one hand, each component can be classified into more than 150 different categories. And the importance of each category varies depending on the type and family to which the component belongs. This fact makes offering a search or browsing using filters very complicated.

On the other hand, the names and descriptions of entire families of components are very similar, and the difference may be one specific characteristic among hundreds. This matter makes an open search usually inconclusive for the users. They are forced to repeat the search over and over again to find the component they are interested in.

Because of the problems related to categories, facets and open search, navigation through the website was confusing and complicated. This led to a considerable increase in site abandonment rates and consequently low user engagement on the platform.

Solution

Optimization of the faceted classification system

The main improvement in the search system was thanks to the development and improvement of the facet system. Each component of the platform could belong to more than 100 different categories. In a faceted search interface, it is redundant to offer the user more than a limited number of facets, each with a limited number of elements. If so, the faceted search system is unusable and therefore inefficient.

The aim was to dynamically offer the users only those facets that were most relevant to them. The most relevant facets in each search are calculated using clustering algorithms implemented in TensorFlow based on the similarity between facets and components. Based on the components returned by the search, the facets that are closest to these components are calculated based on the number of occurrences and the distribution of components within that facet. From there, the facets that best represent the components returned by the search would be searched for.
 

Component recommendation system

In order to make it easy for the user to explore and compare similar components, we set out to provide a list of related components. The problem is that a system of related components based only on similar categories or keywords did not provide the user with valuable information due to the extensive taxonomy tree and the similarities in terms of keywords between the components. So we set up a recommendation system by applying clustering techniques in Tensorflow. This system consists of providing suggestions based on user navigation similar to the current one. The system can learn and feed on the navigation of users who visited the site before to make navigation easier for future users.
 

Results

Once the systems were implemented, we conducted a comprehensive measurement of some specific KPIs and observed that the user engagement rate had increased by an average of 25%.

On the other hand, facets started to be used 30% more in total searches, increasing the number of successful searches per user.

1 in 5 users also made use of the recommendation system (clicked on some of the associated components).