Understanding the Query Intent Behind Searches
A taxonomy of web search (pdf) by Andrea Broder is about differences between informational, transactional, and navigational queries. Those are good to know about when optimizing pages for query terms.
Do you expect to be teaching people about a concept or topic? If so, you are treating their query as if it is an informational one.
Are you creating a page that aims at selling goods or a service? Then you are treating their query as if it is transactional, and enabling them to buy something or book something.
A query used to help someone find a page they are aware of because they have seen it before, or expect it to exist, and they expect a particular page returned to them is referred to as a navigational query. When someone searches for the name of a particular product or brand, they will be satisfied with a home page for that product or brand.
A site I started doing SEO on as an inhouse SEO helped people incorporate businesses. That page ranked well for terms such as “incorporate in Delaware.” It got more conversions when it was the second-ranked site for that term after the website for the Delaware Division of Corporations, which was an informational only site, which did not offer visitors the chance to incorporate their businesses in Delaware.
People would see the Division of Corporations site in search results, visit it and learn about the process, and then return to the search results, and visit the transactional sites, like mine, where they could incorporate. It was an instance where being the highest-ranking site wasn’t an advantage. Understanding the intent behind such queries helps explain why that happened.
More About Query Intent
Another paper about the intent behind searches and queries is A Simple Model for Classifying
Web Queries by User Intent (pdf) by D. Irazú Hernández, Parth Gupta, Paolo Rosso, and Martha Rocha. It makes some interesting statements about query intent that are worth repeating, including this one:
Query Classification based on user intent aims to classify queries about the need behind the queries. Jansen and Booth , define user intent as the expression of an affective, cognitive, or situational goal in an interaction with a Web Search Engine. Query Classification based on user intent is different from traditional text classification because of mainly two issues : first, web queries are usually very short; second, many queries are ambiguous and it is common than a query belongs to multiple categories. For example, for the query “opera theatre tickets”, it is difficult to identify if the user wants to know the website or to buy tickets to attend the event. Most of the efforts have usually involved small quantities of queries manually classified.
While that paper provides some ideas and approaches to a better understanding of query intent, I liked how it framed the problem that it was aiming to address. A patent application from Google that was published this past week describes how the search engine might attempt to understand the context of searches to better understand the intent of a searcher behind their queries.
Google’s Patent on Predicting Query Intent Based Upon Context
It isn’t a new concept, that we can associate an intent with a query, and that Google might try to do so to understand what a person is searching for when then type a handful of words into a search box. What does seem to be new about this patent is how much effort Google may go through to try to understand query intent. When someone types the word “pizza” into Google at lunchtime, we can guess that they may be interested in eating and maybe looking for a place to either pick up some pizza at, or that might deliver to them.
It’s possibly less likely that they are looking for the history of pizza throughout time (but they could be.) How much context might be needed to make such assumptions good ones for most searches?
Like most patents, this new one from Google tells us about the problem it is intended to solve:
If a search query is not narrowly tailored, or if the user does not provide much in the way of additional information beyond the query, a computing device may return too much information; with some of the most interesting or relevant information being difficult for a user to find. The user may experience stress and/or waste valuable time and resources inputting very detailed queries and into a computing device, causing the computing device to execute multiple searches, or sifting through large quantities of search results, to obtain information necessary to accomplish the certain task.
If that searcher is hungry and wants some pizza quickly, their satisfaction with a search engine may increase rapidly if it can identify a nearby location that can deliver tasty pizza.
Using Context to Predict Query Intent
Query intent can be used to enable a search engine to adjust search results returned from the search so information for satisfying the intent is emphasized over other information returned from the search.
The patent gives us an example of someone searching for a movie that they have just purchased tickets for. The system might be able to look at log data, and determine that the searcher had already bought tickets for a future showing of that movie, and may adjust the results so that “movie showtimes are ranked lower than other information (e.g., reviews, memorabilia, trivia, etc.) about the particular movie.”
We are familiar with a search engine collecting information about previous searches that we performed to personalize the results that we might see, but this use of context aimed at reducing stress or avoiding wasting time is different.
The patent application tells us that it may look at such contextual information only after receiving permission from the person using the search engine to analyze that information. The patent provides additional examples of how they might use contextual information that might change the order of search results that they might show a searcher.
Keep in mind that this is still a pending patent application, and it likely hasn’t been implemented yet but could be at some stage in the future and that it is likely that Google is paying attention to context to predict the intent behind queries, and can use things such as time of day, day of the year, and location already.
The patent does tell us that it will limit the use of some information, such as any that might reveal personally identifiable information about a searcher.
This newly published patent application can be found at:
Predicting Intent of a Search for a Particular Context
Publication Number: 20180336200
Publication Date: November 22, 2018
Applicants: Google Inc.
Inventors: Yew Jin Lim, Joseph Linn, Yuling Liang, Carsten Steinebach, Wei Lwun Lu, Dong Hyun Kim, James Kun, Lauren Koepnick and Min Yang
A computing system is described that determines, based on user-initiated actions performed by a group of computing devices, and intent of a search using a particular search query received from a computing device. The computing system adjusts, based on the intent, at least a particular portion of search results obtained from the search using the search query by emphasizing information that satisfies the intent. The computing system sends, to the computing device, an indication of the adjusted search results.
Contextual History Examples
The patent application tells us about some of the different types of contextual information that might be used to predict query intent, including a list of things they refer to as topics of interest that could be found by looking as such things as:
- A user interest graph or some other type of data structure
- Contact information associated with users (e.g., a user’s personal contact information as well as information about a user’s friends, co-workers, social media connections, family, etc.)
- Search histories
- Location histories
- Long and short term tasks
- Calendar information
- Application use histories
- Purchase histories
- Other information
In addition to these personal interest type things, they include other contextual information.
About the operating state of a computing device (and they dig deeply here.):
- Positions of switches
- Battery levels
- Whether a device is plugged into a wall outlet or otherwise operably coupled to another device and/or machine
- User authentication information (e.g., which user is currently authenticated-on or is the current user of the device)
- Whether a device is operating in “airplane” mode, in standby mode, in full-power mode
Other Examples of Contextual Information:
- An acoustic fingerprint
- A video fingerprint
- A location
- A movement trajectory
- A direction
- A speed
- A name of an establishment
- A street address
- A type of place
- A building
- Weather conditions
- Traffic conditions
- A calendar event
- A meeting, or other event associated with a location and/or time
- A webpage address viewed at a particular time
- One or more text entries made in data fields of the webpages at particular times including search or browsing histories
- product purchases made at particular times
- product wish lists
- Product registries
- Audio and/or video accessed by or being broadcast in the presence of the computing device at various locations and times
- Television or cable/satellite broadcasts accessed by or being broadcast in the presence the computing device at various locations and times
- Information about other services accessed by the computing device at various locations and times.
We see other types of information appearing in this patent, like Mobile location history, which I have written about in some detail before.
The patent application tells us about how it may index such contexts to enable it to search for information that might be relevant to a query that may cause it to supplement or modify the search results of such a query.
It also tells us that it might maintain search histories, separate from contextual histories that the search engine may maintain about searches and devices that searches are run on. Google may not interrogate a searcher to their intent behind a search, but it is sounding like they may be able to learn a lot about the people and the machines behind a search to give them the ability to predict what to show in search results.
Machine Learning to Determine Query Intent
A Google patent tells us the process behind it involves collecting a lot of information. It may make predictions after filtering out unnecessary information so that it can “define a narrow context so a true intent of the search query can be inferred.”
The patent tells us about how it might use deep-learning to help in this task as well:
Prediction module may execute a machine-learning model (e.g., a deep-learning model) that receives as inputs: a search query (or portion of a search query) and a current context received from context module. The machine-learning model may generate as output, an indication (such as a label or other identifier) of an intent of a search using the search query for the current context.
The patent provides details about predefined intents (like travel), and intent scores, and how user data may train the machine learning about query intent involved.
This is a patent application, and Google may not have implemented creating such contextual histories and adjusting search results based upon them. Also, keep in mind that the patent says they will likely ask for permission before they analyze such contextual data.
To repeat the query intent based upon context example I shared above:
- You purchase movie tickets
- You perform a search for that movie
- The search engine notices your purchase history
- Instead of showing you movie times for other showings of the same movie, your search results may be adjusted to show you trivia and memorabilia and news about the movie
While Google is deeply tracking Search History and Location history, this understanding of query intent and use of contextual history could potentially change rankings much more than search personalization does.