Why contextual search is a revolution in getting better customer answers from CX data.
With another strong update to the search capability in the Insights Hub, finding the customer answer inside and across your experience data just got a lot easier, faster and more precise. Read below to learn how contextual search will increase the value and use of CX data in companies.
Using neural networks to allow you to search contextually in the Insights Hub
You can search for quotes or observations in a study or across studies on the Insights Hub and use the results to create insights. Now, with the help of state of the art neural networks known as ‘transformers’, you can get better search results based on context. Last year, Google said they would be using a similar model to improve 1 in every 10 searches in English.
Neural nets for language understanding
The magic behind the groundbreaking idea of bidirectional transformers is to train deep bidirectional representations from unlabeled text by holding out words and sentences then trying to predict them by looking at the rest of the sentence. The result is a model with a deep understanding of language. In the Insights Hub, we use a model like this to understand context when searching.
Most search engines are lexical in the sense that they look for a literal match of the query words without understanding the exact query. When searching the internet or a very large database, you’re almost sure you’ll find a match since you have so many possible matches.
Notice in the screenshot for the keyword search that the matches are based on exact matches of the words in the query, such as “financial” and “industry”.
Searching contextually is when the search engine tries to understand the query instead of finding literal matches of words. It matches the search with text that has a similar contextual meaning.
As an example, consider that you have a small data set and you search for the word “expensive”. You first search keywords and you don’t get any hits on the exact word ‘expensive’ in your data set. But by using the context you could get hits on synonyms or semantically similar words such as “costly”, “high priced”, “pricey” etc. In this way, you get a much higher chance of finding relevant quotes for your search query.
If you write a longer search query, you can improve the result and help the model understand the query. It is, therefore, best if your search queries are full sentences. As an example, consider the different meanings of the word bank in this sentence; “after stealing money from the bank vault the bank robber was fishing on the Mississippi riverbank“. Bank as a word by itself can have many meanings, but as we add more words to the sentence we get the context and so does the search engine.
Example of Contextual Search
Notice in the screenshot for the contextual search that the matches are based on the context of the question and not the exact matches of words in the query.
In addition, our model for contextual search is multilingual and you can, therefore, search in English and get results in other local languages or visa versa. This will enable you to search across studies and create insights across markets. See Zwipe case for example.
For more information about the Insights Hub, click here.