Course:LIBR557/2020WT2/query suggestion
Query Suggestion
Query Suggestion is a tool that displays a list of possible queries based on prior users' behaviour that a user can select from as they type (Algolia, 2021). It is one of the most popular query modification techniques1 that reduces ambiguity and inaccuracy in information retrieval and helps the effectiveness and efficiency of the result (Ooi, Ma, Qin & Liew, 2015). Most business search engines such as Google, Bing, and Yahoo are adopting this method to provide a better user experience.
Suggesting Queries at Each Keystroke
One regular way to implement query suggestion is by auto-filling. When a user is typing, a menu will appear with a list of queries predicting what they want to type in (Hearst, 2009). For example, when a user types in “dog”, the menu may show “dog breeds”, “dog names”, or “dog adoption Vancouver” (See Fig.1 for another example). Then, the user can directly select the question they want to ask without further entering the whole query. Query suggestion is dynamic at every keystroke and may change as a user types in more content.
Query Suggestion with Results
In some cases, search engines not only predict possible queries, but also present possible results related to the queries. Hence, users no longer need to go to the results page to see the answer they want (see Fig.2 for example).
Benefits
Due to different levels of cognition and personal habits, queries submitted by users are usually irregular and difficult for search engines to identify the real user needs from their short keywords entered, not to mention returning accurate query results. The technology of query suggestion combines the efforts of humans and algorithms and provides an easier way to formulate quality queries with a single click. Benefits of query suggestion include but not limited to :
- Reducing possible vocabulary problem including misspelling;
- Increase the precision of search sessions for users from all backgrounds (Lopes & Ribeiro, 2018);
- Guaranteeing return results (Kato et al., 2013).
Methods2
Click-through Based Query Suggestion
After a user submits a query, the search engine records their click of the URLs. When two queries share a large number of clicked URLs, they will be identified as similar queries.
Clustering Based Query Suggestion
Clustering Based Query Suggestion clusters queries into groups based on clicked URLs. For a given query Q, the algorithm will find the cluster to which it belongs, and the cluster will be presented as the query suggestion.
Bipartite Graph Based Query Suggestion
In the bipartite graph method, each query forms a pair: <query, URL>. Bipartite graphs are formed by margining all the pairs, and each query and URL become nodes from two sides on the graph. By looking for disjoint similar query sets and disjoint URLs, search engines could find different expressions of queries with similar requirements and pages containing similar information.
Session Based Query Suggestion
Session Based Query Suggestion predicts user behaviour using query sequences. Evidence has shown that many users only search for one topic only in one search session. In most cases, a single search session contains multiple queries which are different expressions of the same topic entered by the same person. Based on this feature, session based query suggestion is developed. Broadly speaking, there are dominantly two types of session-based methods: Adjacency Based Query Suggestion and Query Flow Graph Based query suggestion.
Adjacency Based Query Suggestion
When many users of a search engine submitted Q2 after Q1 in one search session, Q1 and Q2 will be identified adjacent. Then, when a user enters Q1 in their session, the search engine will suggest Q2.
Query Flow Graph Based Query Suggestion
Query Flow Graph Based Query Suggestion uses a graph to represent information. Each node on the graph stands for a query. Each path symbolizes a search behaviour. Once a transition from Node A and Node B is detected, the weight of the path increase. Search engines using the query flow based method suggest queries with the heaviest weight when a user type in a query.
Challenges
Coverage of Query Suggestion
Similar to the collaborative filtering system, suggestions made by the search engines are based on previous user interactions. Therefore, one key challenge of query suggestion for the search engines is its coverage. When there are rarely users searching for a certain topic, how to provide them with high-quality query suggestions? One possible solution proposed by Jain et al. (2011) is to synthesize original queries by tokenizing and relaxing and append context from them to provide suggestions beyond the search log.
The Drift of Query Intention
While some queries were common in the past, users' search intentions change over time. How to combine current events to provide users with the most accurate advice is a new challenge (Meng, 2014). In a given period of time, the relationship between two queries cannot be judged only by the search frequency of past users. Meanwhile, factors as the user’s browsing behaviour, location, demographic features, should also be considered (White, Bilenko & Cucerzan, 2007).
Biases in Autocomplete Suggestions
Although search engine companies argue that suggestions are automatically formed based on users’ behaviour, arguments have been made that companies are obliged to prevent presenting “inappropriate” suggestions that may increase discrimination and hate. One example of biased suggestion is that when a person searched “three black teenagers” on Google in 2016, it shows mugshots; however, when searching “three white teenagers” it shows “wholesome teens” (Nobel, 2018).
Suggestions for Professional Searches
In professional search environments such as legal or medical searches, users often search for the same topic multiple times and read more search results, where boolean queries are regarded as a must for easier manipulation and better self-documentation. At the same time, the question of how to provide search suggestions for professionals has become a new question. To address this issue, for example, Kim, Seo & Croft (2011) has proposed a new technique where boolean queries are seen as sequences of terms associated by conjunctions (AND) and negations (NOT).
Bibliography
- Algolia. (2021). Query Suggestions. Retrieved from https://www.algolia.com/doc/guides/building-search-ui/ui-and-ux-patterns/query-suggestions/js/
- Hearst, M. (2009). The design of search user interfaces. In Search user interfaces (pp. 11–13). Cambridge, United Kingdom: Cambridge University Press
- Jain, A., Ozertem, U., & Velipasaoglu, E. (2011). Synthesizing high utility suggestions for rare web search queries. Proceedings Of The 34Th International ACM SIGIR Conference On Research And Development In Information - SIGIR '11, 805-814. doi: 10.1145/2009916.2010024
- Kato, M. P., Sakai, T., & Tanaka, K. (2013). When do people use query suggestion? A query suggestion log analysis. Information Retrieval, 16(6), 725–746. https://doi.org/10.1007/s10791-012-9216-x
- Kim, Y., Seo, J., & Croft, W. (2011). Automatic boolean query suggestion for professional search. SIGIR '11: Proceedings Of The 34Th International ACM SIGIR Conference On Research And Development In Information, 825-834. https://dl-acm-org.ezproxy.library.ubc.ca/doi/10.1145/2009916.2010026
- Lopes, C., & Ribeiro, C. (2018). Effects of Language and Terminology of Query Suggestions on the Precision of Health Searches. Lecture Notes In Computer Science, 101-111. doi: 10.1007/978-3-319-98932-7_9
- Meng, L. (2014). A Survey on Query Suggestion 1. International Journal of Hybrid Information Technology, 7(6), 43–56. https://doi.org/10.14257/ijhit.2014.7.6.04
- Noble, S. (2018). Algorithms of oppression: How Search Engines Reinforce Racism (pp. 80-81). New York: NYU Press.
- Ooi, J., Ma, X., Qin, H., & Liew, S. (2015). A survey of query expansion, query suggestion and query refinement techniques. 4Th International Conference On Software Engineering And Computer Systems (ICSECS). doi: 10.1109/icsecs.2015.7333094
- White, R., Bilenko, M., & Cucerzan, S. (2007). Studying the use of popular destinations to enhance web search interaction. Proceedings Of The 30Th Annual International ACM SIGIR Conference On Research And Development In Information Retrieval - SIGIR '07. doi: 10.1145/1277741.1277771
Footnotes
1. The other popular query modification techniques include query expansion, query refinement, etc.
2. The classification of query suggestion methods refers to the paper of Meng (2014).