Google just lately filed a brand new patent for a method to offer search outcomes primarily based on a consumer’s searching and e mail historical past. The patent outlines a brand new technique to search inside the context of a search engine, inside an e mail interface, and thru a voice-based assistant (referred to within the patent as a voice-based dialog system).
An issue that many individuals have is that they will bear in mind what they noticed however they will’t bear in mind the place they noticed it or how they discovered it. The brand new patent, titled Producing Question Solutions From A Person’s Historical past, solves that drawback by serving to individuals discover info they’ve beforehand seen inside a webpage or an e mail by enabling them to ask for what they’re searching for utilizing on a regular basis language equivalent to “What was that article I learn final week about chess?”
The issue the invention solves is that conventional serps don’t allow customers to simply search their very own searching or e mail historical past utilizing pure language. The invention works by taking a consumer’s spoken or typed query, recognizing that the query is asking for beforehand seen content material, after which retrieving search outcomes from the consumer’s private historical past (equivalent to their browser historical past or emails). With the intention to accomplish this it makes use of filters like date, matter, or gadget used.
What’s novel in regards to the invention is the system’s capability to grasp imprecise or fuzzy pure language queries and match them to a consumer’s particular previous interactions, together with exhibiting the model of a web page because it seemed when the consumer initially noticed it (a cached model of the online web page).
Question Classification (Intent) And Filtering
Question Classification
The system first determines whether or not the intent of the consumer’s spoken or typed question is to retrieve beforehand accessed info. This course of is named question classification and includes analyzing the phrasing of the question to detect the intent. The system compares components of the question to identified patterns related to history-seeking questions and makes use of methods like semantic evaluation and similarity thresholds to establish if the consumer’s intent is to hunt one thing they’d seen earlier than, even when the wording is imprecise or conversational.
The similarity threshold is an fascinating a part of the invention as a result of it compares what the consumer is saying or typing to identified history-seeking phrases to see if they’re comparable. It’s not searching for a precise match however slightly a detailed match.
Filtering
The subsequent half is filtering, and it occurs after the system has recognized the history-seeking intent. It then applies filters equivalent to the subject, time, or gadget to restrict the search to content material from the consumer’s private historical past that matches these standards.
The time filter is a technique to constrain the search to inside a particular timeframe that’s talked about or implied within the search question. This helps the system slender down the search outcomes to what the consumer is looking for. So if a consumer speaks phrases like “final week” or “a couple of days in the past” then it is aware of to limit the question to these respective time frames.
An fascinating high quality of the time filter is that it’s utilized with a degree of fuzziness, which suggests it’s not precise. So when an individual asks the voice assistant to search out one thing from the previous week it gained’t do a literal search of the previous seven days however will increase it to an extended time period.
The patent describes the fuzzy high quality of the time filter:
“For instance, the browser historical past assortment… might embrace a listing of net pages that had been accessed by the consumer. The search engine… might get hold of paperwork from the index… primarily based on the filters from the formatted question.
For instance, if the formatted question… features a date filter (e.g., “final week”) and a subject filter (e.g., “chess story”), the search engine… might retrieve solely paperwork from the gathering… that fulfill these filters, i.e., paperwork that the consumer accessed within the earlier week that relate to a “chess story.”
On this instance, the search engine… might apply fuzzy time ranges to the “final week” filter to account for inaccuracies in human reminiscence. Specifically, whereas “final week” actually refers back to the seven calendar days of the earlier week, the search engine… might seek for paperwork over a wider vary, e.g., anytime up to now two weeks.”
As soon as a question is assessed as asking for one thing that was beforehand seen, the system identifies particulars within the consumer’s phrasing which might be indicative of matter, date or time, supply, gadget, sender, or location and makes use of them as filters to go looking the consumer’s private historical past.
Every filter helps slender the scope of the search to match what the consumer is making an attempt to recall: for instance, a subject filter (“turkey recipe”) targets the topic of the content material; a time filter (“final week”) restricts outcomes to when it was accessed; a supply filter (“WhiteHouse.gov”) limits the search to particular web sites; a tool filter (e.g., “on my cellphone”) additional restricts the search outcomes from a sure gadget; a sender filter (“from grandma”) helps find emails or shared content material; and a location filter (e.g., “at work”) restricts outcomes to these accessed in a specific bodily place.
By combining these context-sensitive filters, the system mimics the way in which individuals naturally bear in mind content material with a view to assist customers retrieve precisely what they’re searching for, even when their question is imprecise or incomplete.
Scope of Search: What Is Searched
The subsequent a part of the patent is about determining the scope of what will be searched, which is proscribed to predefined sources equivalent to browser historical past, cached variations of net pages, or emails. So, slightly than looking out the complete net, the system focuses solely on the consumer’s private historical past, making the outcomes extra related to what the consumer is making an attempt to recall.
Cached Variations of Beforehand Considered Content material
One other fascinating characteristic described within the patent is net web page caching. Caching refers to saving a replica of an online web page because it appeared when the consumer initially seen it. This permits the system to indicate the consumer that particular model of the web page in search outcomes, slightly than the present model, which can have modified or been eliminated.
The cached model acts like a snapshot in time, making it simpler for the consumer to acknowledge or bear in mind the content material they’re searching for. That is particularly helpful when the consumer doesn’t bear in mind exact particulars just like the title of the web page or the place they discovered it, however would acknowledge it in the event that they noticed it once more. By exhibiting the model that the consumer really noticed, the system makes the search expertise extra aligned with how individuals bear in mind issues.
Potential Purposes Of The Patent Invention
The system described within the patent may be utilized in a number of real-world contexts the place customers might wish to retrieve content material they’ve beforehand seen:
Search Engines
The patent refers a number of occasions to the usage of this system within the context of a search engine that retrieves outcomes not from the general public net, however from the consumer’s private historical past, equivalent to beforehand visited net pages and emails. Whereas the system is designed to go looking solely content material the consumer has beforehand accessed, the patent notes that some implementations may embrace extra paperwork related to the question, even when the consumer hasn’t seen them earlier than.
Electronic mail Shoppers
The system treats beforehand accessed emails as a part of the searchable historical past. For instance, it might return an previous e mail like “Grandma’s turkey meatballs” primarily based on imprecise, pure language queries.
Voice Assistants
The patent contains examples of “a voice-based search” the place customers converse conversational queries like “I’m searching for a turkey recipe I learn on my cellphone.” The system handles speech recognition and interprets intent to retrieve related outcomes from private historical past.
Learn the complete patent right here:
Producing question solutions from a consumer’s historical past
Featured Picture by Shutterstock/JHVEPhoto
