The very first time a user inputs a key stroke in the search box his search experience can be enhanced by either providing query suggestions or performing type ahead queries and showing the results.
Google and Bing are examples of search systems providing query suggestions.
Below are the query suggestions presented by Google while searching for “Designing with the Mind in Mind”
For the same query the user needs to type more characters in Bing to get the suggestion.
Linkedin and Facebook are examples of search systems performing type ahead queries.
Here is an example of type ahead search done on Linkedin while searching for “Yelp”
Providing search suggestions is more appropriate for systems in which users engage in textual or keyword based queries, while type ahead queries work better when users are searching for well known entities.
Benefits of Search Suggest Systems
Some of the benefits that users get from suggested queries are
- Spelling mistakes are avoided. Words like “Ratatouille” are almost impossible to spell correctly. Most search systems are unforgiving of mistakes made while entering the query terms. By providing query suggestions we greatly improve the users chance of finding what he is looking for by eliminating a large class of spelling mistakes.
- Fewer keys need to be typed. Most users do not touch type. Every key press involves significant cognitive load on the user. By providing suggestions that can be easily selected we reduce the efforts that users must take to use the search system.
- Suggestions provide immediate feedback in terms of what queries are possible. Query suggestions also provide immediate feedback in terms of what kind of queries the search system is capable of serving. The presence of a suggestion indicates that the search system should provide a useful response to the query. The absence of a suggestion could often indicate that the system may not contain appropriate responses for the query.
Goals for search suggest systems.
- Users should perceive that they are receiving the query suggestion
From the above user must see the query suggestions within 150ms of pressing a key to perceive that they are instantaneous.
- Suggestions should be ordered by likelihood of them being entered
by the user.
Most search suggest systems are built by performing analysis of logs of the search system. Queries are scored based on how frequently they are entered and if they resulted in clicks on the search results. High scoring queries become the candidates for suggestions. In addition the ranking of the suggestions can be personalized for a particular user. The rankings can also be modified based on the category of the suggestion. For example for a library ILS system suggestions for book titles can be ranked higher than suggestions for DVD titles during weekdays while on weekends DVD titles could be ranked higher. results. While suggestions should be presented in the order of their score there might be a benefit to re-ordering them so that if a suggestion is the first word of another suggestion but it ranks lower then it should still be presented first.
For a user to believe that the search suggestion systems is correctly guiding him it is important that the system be complete in terms of the suggestions that it provides. If a query has the potential of resulting in a search result with a high click through rate then that query should be provided as a suggestion when its prefix has been typed by the user. Here is an example of a search system that fails to do this. The user is searching for the book "Lady Tasting Tea" on Ebay. After typing "lady ta" he stops receiving suggestions.
However when he completes the query he does find that Ebay sells the book.
The query suggestions system should not provides suggestions which result in zero search results. Such suggestions reduce the users belief that the search auto complete system is correctly guiding his search exploration. It is indeed very rare to find a suggest system in production that does this.