Case Study

Product Reference Search Tool with Attention Layer


An insurance carrier had implemented an in-house search tool to facilitate employee access to its product reference manuals. However, administrators noted that users rarely typed full questions using English language syntax. The data science research team determined that adding an attention layer could convert these partial questions into the correct syntax, thereby increasing the performance of the search tool.


Financial Services / Insurance

Data Type


Project Duration

1 Month




Data Associates were provided with raw syntax and 20 machine-generated questions for each search query. They selected all the machine-generated questions that used the correct syntax and then created 3 additional questions using the full and correct syntax.


Selecting the correct machine-generated questions and adding additional examples strengthened the attention layer so that it could accept partial / poorly structured user queries and consistently output strings which the search tool could correctly interpret – thereby increasing the usage of the product reference manuals.

Download Case Study