Generating narrative query response from search-based auto-suggestion query using generative language model

Abstract

The present disclosure describes a query gateway system that provides an efficient and flexible framework for providing context-preserved auto-suggestion queries from an auto-suggestion query system (e.g., a search engine query experience) to a generative language model system (e.g., an AI chat experience). For example, a query gateway system establishes a framework to automatically provide context-preserved queries to a generative language model system using features and services of an automatically suggested query system and using a separate user interface that does not interfere with user navigation or requires manual replication of user input. Additionally, the query gateway system incorporates additional enhancements, including an AI chat eligibility model and a query reformulated model, to improve the computational efficiency and accuracy of the AI chat system.

Description

Background Technology

[0001] Search engine services have significantly enhanced the ability to explore websites across vast areas of the internet. Recently, advanced chat services utilizing artificial intelligence (AI) have emerged, known as generative language models (GLMs), including large language models (LLMs). These GLMs employ machine learning models to generate narrative-based responses based on user queries. While these services typically operate independently, some service providers have begun incorporating links between them. However, a framework for context-preserving integration between services remains lacking. Furthermore, while certain features can enhance one service, there are currently no suitable mechanisms to leverage these advancements to improve another. Attached Figure Description

[0002] The following detailed description provides specific and detailed implementations accompanied by the accompanying drawings. Additionally, each figure listed below corresponds to one or more implementations discussed in this disclosure.

[0003] Figure 1 The diagram illustrates an overview example of a query gateway system that implements context-preserving autosuggestion queries, facilitating transitions from autosuggestion query systems to generative language model systems.

[0004] Figure 2 The diagram illustrates the system environment for implementing the query gateway system.

[0005] Figure 3 The diagram illustrates a flowchart example using auto-suggestion queries, which leverages both an auto-suggestion query system and a generative language model system.

[0006] Figures 4A to 4B The illustration shows an example of a graphical user interface that automatically provides context-preserving autosuggestion queries from the autosuggestion query system user interface to the generative language model system user interface.

[0007] Figure 5 The diagram illustrates an example block diagram for providing a reformulated autosuggestion query based on an autosuggestion query system to a generative language model system.

[0008] Figure 6 The diagram illustrates an example flowchart for generating reformulated autosuggested queries from autosuggested queries.

[0009] Figures 7A to 7B The illustration shows an example graphical user interface that uses a generative language model from multiple autosuggestion queries to generate a narrative-based response.

[0010] Figure 8The illustration shows a series of example actions in a computer implementation of a method for generating narrative query responses using a generative language model.

[0011] Figure 9 The diagram illustrates the components included within the example computer system. Detailed Implementation

[0012] This disclosure describes a query gateway system that provides an efficient and flexible framework for delivering context-preserving autosuggestion queries from an autosuggestion query system to a generative language model system. Specifically, the query gateway system establishes a framework to leverage features and services of the autosuggestion query system, including relevant autosuggestion queries, to enhance queries provided to the generative language model system. Additionally, the query gateway system incorporates additional enhancements such as AI chat qualification models and query reformulation models to improve the computational efficiency and accuracy of the AI ​​chat system.

[0013] In some implementations, query gateway systems combine the autosuggestion features of web-based search experiences with AI chat-based experiences. For example, a query gateway system selectively generates and provides additional graphical elements within an autosuggestion query interface (e.g., a web-based search experience) to initiate a transition to a generative language model system interface (e.g., an AI chat-based experience). Additionally, when transitioning between search and chat experiences, the query gateway system facilitates the preservation of the context of the autosuggestion query and, in many cases, provides additional context to queries offered to the AI ​​chat system. Furthermore, the query gateway system provides a flexible framework that minimizes navigation disruption during transitions between search systems.

[0014] To illustrate, in some implementations, the query gateway system utilizes a generative language model to generate narrative query responses, such as AI chat responses. For example, when text input (e.g., a text prefix) corresponding to a search query (e.g., web search) is detected, the query gateway system identifies one or more autosuggested queries. Additionally, in some instances, the query gateway system generates a generative language model (GLM) eligibility cache for the autosuggested queries. Then, based on determining that the autosuggested queries are eligible for use with a generative language model (e.g., an AI chat model) using the generative language model eligibility cache, the query gateway system provides generative language model elements for display immediately adjacent to the autosuggested queries within a first user interface. Upon detecting the selection of a generative language model element, the query gateway system generates a reformulated autosuggested query from the original autosuggested query. Further, the query gateway system provides the reformulated autosuggested query to the generative language model for display in a second user interface separate from the first user interface.

[0015] As described in this document, the query gateway system delivers several significant technical benefits in terms of computational efficiency, accuracy, and flexibility compared to existing systems. Furthermore, the query gateway system provides several practical applications that address problems through a presentation framework that preserves the context of one or more autosuggested queries from an in-query autosuggestion system, which are typically reformulated and presented to a generative language model system, thereby generating several benefits, as further demonstrated below.

[0016] To illustrate, the query gateway system adds an AI chat element (i.e., a GLM element) to the autosuggested query pane of the search user interface for each eligible autosuggested query. Then, upon selection of an AI chat element immediately adjacent to the first autosuggested query, the query gateway system opens a separate AI chat user interface (i.e., the GLM interface), which includes a first narrative-based response from the GLM based on the first autosuggested query. Upon detecting the selection of another AI chat element immediately adjacent to the second autosuggested query (because the autosuggested query remains within the search user interface even when an AI chat element is selected), the query gateway system opens another separate AI chat user interface, which includes a second narrative-based response from the GLM based on the second autosuggested query.

[0017] By incorporating generative language model elements (e.g., an AI chat button) into the user interface that triggers the transition to the GLM system, the query gateway system provides an improved user interface that significantly reduces the number of navigation steps currently required to switch from the autosuggested query system to the GLM system. To illustrate, while existing systems provide autosuggested queries in response to user-inputted text prefixes, these queries only allow the user to perform a corresponding web search using the selected autosuggested query. If the user wants to switch to an AI chat-based experience, they need to navigate to the GLM user interface and manually re-enter the autosuggested query. Occasionally, after performing a traditional web search, the existing system offers an option that, when selected, forwards the autosuggested query to the GLM system for further processing. In contrast, the query gateway system allows users to automatically transition from the autosuggested query to the GLM interface without requiring them to manually navigate to the GLM interface or re-enter the autosuggested query.

[0018] This benefit is further amplified when multiple auto-suggestion queries are selected to be provided to the GLM system. For example, using an existing system, if a user expects to receive responses from multiple auto-suggestion queries based on a text input prefix, the user might need to re-enter the text input prefix for each selected auto-suggestion query and then copy each selected auto-suggestion query from the auto-suggestion query system to the GLM system. Each time, the user needs to manually switch between each system interface, typically reloading the system's user interface and re-entering the text input prefix and / or auto-suggestion query. In contrast, the query gateway system allows the user to enter a text input prefix once and select the AI ​​chat element corresponding to the multiple auto-suggestion queries provided, without needing to leave or reload the auto-suggestion query system interface. Furthermore, the query gateway system allows the user to access different instances of the GLM system corresponding to each selected AI chat element, where the corresponding auto-suggestion query is automatically loaded into the GLM system.

[0019] By leveraging generative language model elements with autosuggested queries, the query gateway system also improves overall accuracy by preserving the context of the autosuggested queries. For example, the query gateway system automatically provides autosuggested queries to the GLM system without requiring the user to manually select, copy, and / or re-enter the autosuggested queries from the autosuggested query search system to the GLM system. In this way, the query gateway system ensures that context is accurately preserved during the transformation.

[0020] Furthermore, the navigational benefits provided by generative language model elements become even more significant when computing devices have smaller displays. For example, devices with smaller displays may struggle to navigate between interfaces such as browser tabs. Additionally, users need to manually copy and paste or re-enter autosuggested queries, making navigation between multiple interfaces even more difficult. A query gateway system instead provides users with the ability to easily select one or more generative language model elements (e.g., AI chat elements) that offer autosuggested queries, and the query gateway system automatically forwards and loads the corresponding autosuggested queries into the GLM system in new or separate interfaces.

[0021] Additionally, in several instances, the query gateway system provides further improved accuracy by leveraging reformulated autosuggested queries. For example, in these implementations, the query gateway system utilizes autosuggested queries to generate new, reformulated autosuggested queries that are more suitable for the input parameters of the GLM system. In this way, the reformulated autosuggested queries enable the GLM system to generate better and more accurate responses than autosuggested queries that are not suitable for the GLM system.

[0022] Furthermore, by leveraging a generative language model eligibility cache or a reformulated query cache, the query gateway system reduces the number of computational operations required. For example, in some implementations, when an autosuggestion query is selected, the query gateway system determines whether that autosuggestion query is associated with a reformulated autosuggestion query. If a reformulated autosuggestion query for the autosuggestion query exists in the generative language model eligibility cache, the query gateway system does not need to reprocess the autosuggestion query to generate a new reformulated autosuggestion query. If a new reformulated autosuggestion query needs to be generated, the query gateway system uses a lightweight language model to determine the reformulated autosuggestion query in real time. Additionally, the query gateway system stores the newly generated reformulated autosuggestion queries in the generative language model eligibility cache to prevent future regeneration of reformulated autosuggestion queries.

[0023] As a further example, in some implementations, the query gateway system allows the simultaneous selection of multiple autosuggested queries. In these implementations, the query gateway system reformulates the multiple autosuggested queries into a single robust AI chat query to serve it to the GLM system. In this way, the query gateway system provides a significantly more flexible approach than current systems, which require users to manually formulate AI chat queries. Additionally, the query gateway system enables users to receive narrative-based responses from the GLM system with fewer navigation steps. Furthermore, by generating a reformulated autosuggested query from multiple autosuggested queries, the narrative-based responses generated by the GLM system are more accurate.

[0024] This disclosure uses several terms to describe the features and advantages of one or more implementations. For example, the term "automatic suggestion query search system" refers to a search engine system or another type of search system that provides context-sensitive suggestions to a user for a search query, referred to as "automatic suggestion query". For example, an automatic suggestion query search system utilizes a text input field to detect text input corresponding to a search query. When text input is received, "text input" is detected in the form of typed letters, phrases, and / or prefixes. In response, the automatic suggestion query search system provides multiple automatic suggestion queries as input to the search query.

[0025] The term "Generative Language Model System" (GLM system) refers to a high-level computational system that uses natural language processing and machine learning to generate coherent and context-sensitive human-like text, referred to in this document as "narrative-based query response" or "narrative query response." An example of such a model is the Large Language Model (LLM), which has been trained on large datasets and can produce fluent, coherent, and topic-specific text. GLMs have applications in natural language understanding, content generation, text generalization, dialogue systems, language translation, and creative writing assistance. In some instances, GLM systems are referred to as AI chat systems. In other instances, GLMs are called AI chat models, AI chat services, or simply AI chat.

[0026] In this document, a “Generative Language Model Element” (GLM Element) refers to a graphical user interface element connected to an autosuggestion query. In some instances, a GLM Element is referred to as an AI chat element or an AI chat button. In various implementations, one or more GLM Elements are displayed in the user interface of the autosuggestion query search system, such as an autosuggestion pane showing one or more autosuggestion queries for text input. When a GLM Element that corrects to a given autosuggestion query is selected, the query gateway system can trigger a transition to the GLM system based on the given autosuggestion query, as described below.

[0027] Additional details regarding the sample implementation of the query gateway system (i.e., "automatically suggest queries to the GLM gateway system") will be discussed in conjunction with the accompanying figures below. For example, Figure 1 The diagram illustrates an overview example of a query gateway system that implements context-preserving autosuggestion queries, delivering them from an autosuggestion query system to a generative language model system based on one or more implementations.

[0028] like Figure 1 The illustrated series of actions 100 provides an example of a query gateway system that offers a framework for delivering context-preserving autosuggestion queries from an autosuggestion query system to a generative language model system. In various implementations, the query gateway system performs a series of actions 100. In some implementations, an autosuggestion query search system and / or a generative language model (GLM) system coordinated with the query gateway system performs one or more of the actions, or portions thereof.

[0029] Additionally, a series of actions 100 includes an action 102 that identifies an autosuggested query 114 based on text input 110 used for the query search. For example, in response to user-provided text input 110, the query suggestion service determines the autosuggested query 114 from the autosuggested query set 112. In various implementations, the query suggestion service is part of an autosuggested query search system.

[0030] Figure 1 A first user interface 116 related to action 102 is also shown. The first user interface 116 includes a text input field for receiving text input 110 and an autosuggestion pane for displaying autosuggestion queries 114. In various implementations, the autosuggestion query search system displays a portion of the user interface of an autosuggestion query search system, such as a search engine website or product search. Combined Figure 3 and Figure 5 Additional details are provided regarding the generation of automatically suggested queries from text input.

[0031] exist Figure 1 In this context, a series of actions 100 includes actions 104 of generating and displaying AI chat elements for eligible autosuggestion queries. For example, when an autosuggestion query 114 is determined for text input 110, the query gateway system provides one or more autosuggestion queries from autosuggestion query 114 to an AI chat eligibility model (i.e., a GLM eligibility model) with a generative language model cache 120 to determine whether a given autosuggestion query is eligible to be provided to the GLM system. For eligible autosuggestion queries, the query gateway system generates and provides generative language model elements 122 (GLM elements) (such as AI chat elements), which are displayed side-by-side with the autosuggestion query within a first user interface 116. Figure 5 Provide additional details regarding the determination of GLM eligibility.

[0032] As shown, a series of actions 100 includes action 106 of generating a reformulated auto-suggestion query upon detecting the selection of an AI chat element for auto-suggestion queries. For example, in response to detecting the selection of a generative language model element 122 displayed alongside the corresponding auto-suggestion query, the query gateway system sends an auto-suggestion query 124 to the GLM system for processing.

[0033] In some implementations, before providing the autosuggested query 124 to the GLM system, the query gateway system determines a reformulated autosuggested query 126 for autosuggested query 124. For example, the query gateway system uses the query reformulation model to identify and / or generate a reformulated autosuggested query 126 for autosuggested query 124. Typically, the reformulated autosuggested query is comprehensive and detailed (e.g., more robust and verbose), resulting in a better response from the GLM system. Figure 3 , Figure 5 and Figure 6 Additional details regarding the determination of GLM eligibility are provided.

[0034] As shown, a series of actions 100 includes action 108 of providing a reformulated autosuggestion query 126 to the AI ​​chat system in a separate user interface. For example, once the reformulated autosuggestion query 126 is determined, the query gateway system delivers the reformulated autosuggestion query 126 to the GLM system and initiates or causes the GLM system to begin processing the reformulated autosuggestion query 126 as a narrative-based query using a generative language model 130.

[0035] Additionally, as mentioned earlier, in various implementations, the query gateway system generates new user interfaces, such as a second user interface 132, to suit the GLM system. For example, if the first user interface 116 is displayed in a first window or browser tab, the query gateway system creates or generates a new window or browser tab to display the AI ​​chat system. This ensures that when the AI ​​chat system is triggered to provide a narrative query response 134 for the corresponding auto-suggested query, the first user interface 116 and the user's search experience remain uninterrupted. The following is combined with... Figure 3 , Figures 4A to 4B , Figure 5 and Figures 7A to 7B Provides additional details about triggering the GLM system.

[0036] Using a suitable general overview of the query gateway system, additional details are provided regarding the components and elements of the query gateway system (i.e., "automatic query suggestion to GLM gateway system"). For illustration, Figure 2 The diagram illustrates an example system environment for implementing a query gateway system using one or more implementation methods. Although Figure 2 An example layout and configuration of the query gateway system is shown, but other layouts and configurations are possible.

[0037] As shown, Figure 2 A computing environment 200 includes a computing system having a server device 202 and a client device 230 connected via a network 240. The following text, in conjunction with... Figure 9 Further details are provided regarding these and other computing devices. Additionally, Figure 9 Additional details about the network (such as network 240 shown) are also provided.

[0038] As shown, server device 202 includes a content management system 204 that manages digital content hosted and / or accessed by server device 202. For example, content management system 204 facilitates users to perform searches for digital content from websites, databases, or data repositories across the Internet.

[0039] As also shown, the content management system 204 on server device 202 includes a query gateway system 206. In some implementations, the query gateway system 206 is located outside the content management system 204. In various implementations, portions of the query gateway system 206 reside in different components.

[0040] As mentioned above, query gateway system 206 provides a framework for context-preserving gateways between autosuggested queries from an autosuggested query search system and GLM systems (e.g., AI chat systems). In various implementations, query gateway system 206 implements one or more features, such as AI chat qualifications and reformulated autosuggested queries, to provide improved efficiency and accuracy when facilitating transitions between search systems. Additionally, query gateway system 206 utilizes separate user interfaces to simplify navigation and eliminate interruptions when transitioning between search systems. Furthermore, content management system 204 includes autosuggested query system 226 and GLM system 228, which may reside outside of content management system 204 and / or on a different computing device than server device 202 (e.g., GLM system 228 resides on a different cloud computing system).

[0041] As shown, query gateway system 206 includes various components and elements implemented in hardware and / or software. For example, query gateway system 206 includes an autosuggested query manager 210 that communicates with autosuggested query system 226 to generate, identify, refine, and determine autosuggested queries for text input corresponding to a user's search query. Furthermore, query gateway system 206 includes a query eligibility manager 212 that identifies, generates, and / or utilizes a GLM eligibility query 220 to determine when an autosuggested query is eligible to successfully utilize GLM system 228 for operation. Additionally, query gateway system 206 includes a query reformulation manager 214 to determine reformulated queries 222 (e.g., reformulated autosuggested queries) for autosuggested queries to be offered to GLM system 228. Further, query gateway system 206 includes a GLM manager 216 that communicates with GLM system 228 to facilitate the generation of narrative query responses (e.g., AI chat responses) using a GLM model.

[0042] Additionally, query gateway system 206 includes storage manager 218. In various implementations, storage manager 218 stores data corresponding to query gateway system 206. As shown, storage manager 218 includes GLM-qualified query 220, redefined query 222, and machine learning model 224. Machine learning model 224 may include some or all of the following: GLM qualification model (e.g., classifier model), query redefined model (sequence-to-sequence model or lightweight language model), and / or a model corresponding to auto-suggested query system 226 (e.g., auto-suggested query generation model), and / or a model corresponding to GLM system 228 (e.g., LLM).

[0043] Furthermore, computing environment 200 includes client device 230 with client application 232. In various implementations, client device 230 is associated with a user who provides text input and / or expects a more robust, narrative-based query response from the AI ​​chat system. In multiple implementations, the user interacts with server device 202 (e.g., content management system 204 and / or query gateway system 206) to access content and / or services. Computing environment 200 may include any number of client devices.

[0044] As shown, client device 230 includes client application 232. For example, client application 232 is a web browser application, a mobile application, or another type of application for accessing and receiving digital content via the Internet. In some implementations, client device 230 includes a plugin associated with query gateway system 206 that communicates with client application 232 to perform corresponding actions. In some implementations, a portion of query gateway system 206 is integrated into client application 232 to perform corresponding actions.

[0045] Having properly established the query gateway system 206, additional details regarding the various functionalities of the query gateway system 206 will now be described. As noted above, Figure 3 Provide additional details for query gateway system 206. Specifically, Figure 3 The diagram illustrates an example of using autosuggestion queries based on some implementations, which utilize both an autosuggestion query system and a generative language model system.

[0046] As shown, Figure 3This includes a process 300 that provides the user with an initial overview of the query results from an automated suggestion search system or a GLM query system (i.e., a GLM system). For illustration, process 300 includes an action 302 that detects prefixes from text input. For example, the user is presented with a first user interface that includes text input fields for entering characters of text into a search query. In this document, "prefix" refers to a portion of the text input provided by the user before any suggestions are displayed or completion is achieved.

[0047] In response to the detection of a prefix, query suggestion service 304 (e.g., part of an autosuggestion query system) searches the autosuggestion query set 112 to determine relevant or matching autosuggestion queries. For illustration, process 300 includes an action 306 of displaying relevant autosuggestion queries in an autosuggestion pane. For example, upon determining an autosuggestion query, a first user interface updates to display an autosuggestion pane including relevant autosuggestion queries below a text input field.

[0048] At this point, process 300 can branch into two paths. In the first path corresponding to search result 310, process 300 includes action 312 of selecting an auto-suggestion query. For example, the user selects an auto-suggestion query from the auto-suggestion queries. In response, the auto-suggestion query system updates the first user interface to display the search result corresponding to that selection, as shown in action 314.

[0049] In the second path corresponding to the AI ​​chat narrative result 320, the process includes the action 322 of selecting an AI chat element immediately adjacent to the auto-suggested query. As mentioned above, the query gateway system 206 generates and provides AI chat elements (i.e., GLM elements) for one or more auto-suggested queries within the first user interface. The query gateway system 206 then detects when the user selects one of the AI ​​chat elements.

[0050] As shown, process 300 includes query redefinition model 324. In various implementations, when selecting one AI chat element from the AI ​​chat elements, query redefinition model 324 generates a redefined autosuggestion query based on the autosuggestion query corresponding to the selected AI chat element. Query gateway system 206 then provides the redefined autosuggestion query to the AI ​​chat system (i.e., the GLM system).

[0051] Additionally, as shown, process 300 includes action 326 of opening a second user interface of the AI ​​chat system using a redefined autosuggestion query. For example, in conjunction with providing the AI ​​chat system with a redefined autosuggestion query, query gateway system 206 enables the AI ​​chat system to execute the redefined autosuggestion query using an AI chat model (e.g., GLM) and generate a narrative query response. In this way, when the user moves to the second user interface, the narrative query response either waits for the user or is in the process of being generated.

[0052] Figures 4A to 4B A graphical example of the process is provided. Specifically, Figures 4A to 4B The illustration shows an example graphical user interface that automatically provides context-preserving autosuggestion queries from the autosuggestion query system user interface to the generative language model system user interface. As shown, Figures 4A to 4B The device includes a client device 400 (e.g., a laptop computer, desktop computer, smartphone, or tablet computer), which includes a display. The client device 400 can run an operating system (OS) that implements various systems, programs, applications, and services. For example, the client device 400 implements a client application 401 such as a web browser.

[0053] exist Figure 4A In this application, client application 401 includes a first user interface 402, shown as a browser tab. The first user interface corresponds to a search query page and includes an input field 403 where a user can provide text input or another type of input (e.g., audio or image) for a search query. As shown, input field 403 includes text input 404 (e.g., a prefix). Also as shown, the first user interface 402 includes an autosuggestion pane 405 that displays an autosuggested query 406 corresponding to the text input 404.

[0054] Additionally, the first user interface 402 shows an AI chat element 408 immediately adjacent to each of the autosuggestion queries 406. In various implementations, one or more autosuggestion queries in the autosuggestion queries 406 may not include an AI chat element, as discussed below. In some implementations, the first user interface 402 includes pop-ups, additional text, or other guidance indicating the purpose and function of the AI ​​chat element 408. For example, the first user interface 402 includes a guide explaining that the AI ​​chat element 408 will open a new, separate user interface where the corresponding autosuggestion query is used as the initial query with the AI ​​chat system.

[0055] As shown, Figure 4AThis includes using a pointer (such as a mouse or finger) to select one of the auto-suggested queries 406. For example, the selected auto-suggested query 410 is highlighted using different graphics for its AI chat element when selected.

[0056] Upon detecting the selection of the chosen autosuggestion query 410, the query gateway system 206 triggers a transition to the AI ​​chat system. As noted above, in some implementations, the query gateway system 206 reformulates the autosuggestion query before providing it to the AI ​​chat system. In one or more implementations, the query gateway system 206 directly provides the selected autosuggestion query 410 to the AI ​​chat system for processing and / or to provide reformulation instructions.

[0057] Go to Figure 4B The diagram illustrates a client device 400 with a client application 401, which now displays a second user interface 422. Note that the first user interface 402 is still presented, but temporarily in the background. The second user interface 422 corresponds to the AI ​​chat system (i.e., the GLM system) and provides the user with a narrative-based experience, thus offering a more comprehensive query response.

[0058] The second user interface 422 includes a reformulated autosuggestion query 424. For illustration, the selected AI chat element corresponds to the autosuggestion query "Vietnam travel itinerary". Using this as a prompt, the query gateway system 206 generates a more robust and comprehensive query reformulated as a sentence, which is used as input to the AI ​​chat model. In some implementations, the query gateway system 206 provides the corresponding autosuggestion query to the AI ​​chat model with or without additional text input.

[0059] In response, the AI ​​chat model generates and provides a narrative query response 426 to the user. This narrative query response 426 serves as a starting point for dialogue with the user, providing more information about the given topic in a second user interface 422. Furthermore, as illustrated, the AI ​​chat system provides additional prompts to further the conversation.

[0060] As mentioned earlier, when the selected auto-suggestion query 410 is triggered and a second user interface 422 is generated, the client application 401 maintains the first user interface 402. After selecting the selected auto-suggestion query 410, the user can easily navigate to the automatically generated version of the second user interface 422, which includes a description of the query response 426. Additionally, the user can select another AI chat element within the auto-suggestion query 406, which will cause the query gateway system 206 to generate a third user interface that includes another instance of the AI ​​chat system, thereby accurately and automatically applying the context of the new selected AI chat element.

[0061] Furthermore, for each selected AI chat element, the query gateway system 206 can trigger a new instance of the AI ​​chat system along with a context-preserving query in the new user interface. In these implementations, the query gateway system 206 provides the user with quick access to each automatically suggested query of interest with minimal user effort.

[0062] Additionally, when the revised autosuggested query 424 is displayed, the user can quickly navigate back to the first user interface 402, which retains the previously provided text input 404 and autosuggested query 406. This allows the user to proceed to the search results or select another AI chat element to interact with the AI ​​chat system. In effect, the query gateway system 206 facilitates simple, efficient, accurate, and fast navigation between instances of the autosuggested query system and the GLM system with minimal effort required by the user, thereby reducing the likelihood of user errors that often occur when navigating under existing systems.

[0063] Figure 5 The diagram illustrates an example block diagram of an autosuggested query system, based on some implementations, for providing a reformulated autosuggested query system to a generative language model system. Specifically, Figure 5 Framework 500 is shown, and query gateway system 206 utilizes framework 500 to facilitate a smooth gateway between automated suggestion query systems and GLM systems (e.g., AI chat systems). For illustration, framework 500 includes offline framework 502 and online framework 504.

[0064] In various implementations, the offline framework 502 provides functionality and operations for the query gateway system 206 and / or the auto-suggestion query system (such as on periodic scheduling or when a threshold is met) to be performed non-real-time. For illustration, the offline framework 502 includes a query log 506. In various implementations, the query log 506 includes previous search queries performed by the user and may include metadata associated with the search. Additionally, the offline framework 502 illustrates an auto-suggestion data generation pipeline 508 that generates suggestions 510 (e.g., auto-suggestion queries) from previous text input within the query log 506.

[0065] In various implementations, suggestion 510 is stored in an auto-suggestion query cache or another type of data storage medium. Additionally, suggestion 510 is stored in different data structures, such as prefix-based attempts, inverted indexes, or other data structures that utilize auto-suggestion queries to pair text prefixes with corresponding metadata.

[0066] Furthermore, the offline framework 502 includes a classifier model 512. In various implementations, the classifier model 512 employs a machine learning algorithm trained to assign input queries to specific kinds or categories. In this case, the classifier model 512 is trained to classify whether autosuggestion queries are properly suited for the AI ​​chat model. To illustrate, not all autosuggestion queries are well-suited for AI chat systems. For example, autosuggestion queries such as "Bing" or "Microsoft Outlook web email login" might not be suitable for being served to an AI chat system because the search user expects quick access to the requested resource and does not find more information about it.

[0067] In various implementations, the query gateway system 206 can train a classifier model 512 to classify suggestions 510 using previous text input within the query log 506. In some instances, each suggestion in the suggestions 510 is assigned a binary value indicating whether it is suitable for the AI ​​chat system. In some implementations, the classifier model 512 determines the probability that an automated suggestion query (e.g., a suggestion) is suitable for the AI ​​chat system. As mentioned, the query gateway system 206 can periodically update the suggestion classifications when updating the classifier model 512. Also as shown, the classifier model 512 stores the classified suggestions in a chat eligibility cache 522 (i.e., a generative language model eligibility cache), which is discussed further below.

[0068] Similarly, query gateway system 206 can train sequence-to-sequence model 514 (Seq2Seq model) to generate reformulated autosuggested queries (e.g., autosuggested queries) from suggestions 510. In various implementations, a sequence-to-sequence model is a deep learning machine learning language model that maps input sequences to output sequences through encoding and decoding processes. Sequence-to-sequence model 514 can represent different types of language models, such as Large Language Models (LLMs).

[0069] The sequence-to-sequence model 514 receives autosuggested queries and generates reformulated autosuggested queries. In several implementations, the reformulated autosuggested queries are more verbose, robust, and longer than the corresponding autosuggested queries. For example, an autosuggested query might be a string of 3 to 5 words, a string of nouns, and / or a sentence containing bad words. Conversely, a reformulated autosuggested query includes 2 to 3 complete sentences providing context for the query, a query framework, a scenario-based statement, a long-form question, and / or a detailed query request. In effect, the sequence-to-sequence model 514 generates reformulated autosuggested queries that produce more accurate and complete narrative query responses through the AI ​​chat system.

[0070] As shown, the sequence-to-sequence model 514 stores the reformulated auto-suggested queries in the reformulated query cache 534, which is discussed further below. Additionally, the query gateway system 206 periodically updates the sequence-to-sequence model 514 based on the query log 506 and suggestions 510 to generate new reformulated auto-suggested queries and / or update existing reformulated auto-suggested queries.

[0071] In various cases, the classifier model 512 and the sequence-to-sequence model 514 are combined into a single machine learning model. For example, the query gateway system 206 generates a single multi-task model that is jointly trained on the same input data to receive different prompts and provide different corresponding outputs. For example, the multi-task model is an LLM or another type of generative language model.

[0072] In various implementations, the online framework 504 allows the query gateway system 206 to provide users with real-time characteristics of the automated suggestion query system and the GLM system. For illustration, the online framework 504 includes an initial trigger 515 that detects prefixes from the user's text input. For example, when a user visits a website that provides search capabilities, they provide text input into a text input field within a first user interface.

[0073] In response, the autosuggestion query system displays suggestions in an autosuggestion pane, also located in the first user interface, as shown in action 524. Specifically, the autosuggestion query system utilizes a query suggestion service 516 within the online framework 504, which determines autosuggestion queries in real time from prefixes. For example, the query suggestion service 516 accesses suggestions 510 (e.g., an autosuggestion query set) in a suggestion cache and determines relevant autosuggestion queries in real time in response to detecting one or more prefixes. In various implementations, the query suggestion service 516 includes a prefix-based trie (e.g., a prefix tree) service, a non-prefix matching service, a language model-based next-word service, or another service for determining autosuggestion queries from an autosuggestion query set for a given prefix.

[0074] Additionally, the online framework 504 includes an AI chat eligibility model 520 with a chat eligibility cache 522. In various implementations, when the query suggestion service 516 determines an auto-suggestion query for a prefix, the query gateway system 206 determines whether the auto-suggestion query is suitable for the AI ​​chat system.

[0075] To illustrate, in some implementations, query gateway system 206 utilizes AI chat eligibility model 520 to determine whether a given autosuggestion query resides in chat eligibility cache 522. If a match is found in chat eligibility cache 522, query gateway system 206 generates and provides AI chat elements (e.g., action 524) to be displayed alongside the given autosuggestion query in the autosuggestion pane. In some implementations, query gateway system 206 determines a match based on a threshold number or quantity of matching words. In various implementations, AI chat eligibility model 520 is a machine learning model that determines a match based on the proximity of the autosuggestion query to eligible autosuggestion queries in the vector space. If no match is found for a given autosuggestion query, query gateway system 206 does not provide AI chat elements to be displayed alongside the given autosuggestion query in the autosuggestion pane of the first user interface.

[0076] In various implementations, the chat eligibility cache 522 stores only eligible autosuggestion queries. In alternative implementations, the chat eligibility cache 522 stores only unqualified autosuggestion queries that the query gateway system 206 knows are unsuitable for the AI ​​chat system. In some implementations, the chat eligibility cache 522 stores eligible and unqualified autosuggestion queries with positive or negative eligibility indications.

[0077] In some implementations, the online framework 504 excludes or omits the AI ​​chat eligibility model 520. In these implementations, the query gateway system 206 generates an AI chat element for each autosuggestion query provided within the autosuggestion pane.

[0078] As shown, the online framework 504 includes another trigger 526 that detects the selected AI chat element (e.g., a clicked AI chat suggestion). For example, when a user selects an AI chat element, the query gateway system 206 identifies the corresponding autosuggestion query. In several instances, the corresponding autosuggestion query refers to an autosuggestion query that is displayed immediately adjacent to or consistent with the selected AI chat element. The query gateway system 206 then sends the corresponding autosuggestion query to the AI ​​chat system via the query remodeling model 530.

[0079] As shown, the query redefinition model 530 includes a lightweight language model 532 and a redefinition query cache 534. As previously mentioned, the query gateway system 206 utilizes the query redefinition model 530 to generate redefinition autosuggested queries from the input autosuggested queries. For example, the query gateway system 206 provides the corresponding autosuggested queries to the query redefinition model 530, which determines the redefinition autosuggested queries.

[0080] In some implementations, query redefinition model 530 determines whether the corresponding autosuggested query is presented in the redefined query cache 534. If so, query redefinition model 530 identifies the redefined autosuggested query associated with the corresponding autosuggested query from the redefined query cache 534. Otherwise, query redefinition model 530 utilizes a lightweight language model 532 to determine the redefined autosuggested query for the corresponding autosuggested query in real-time (e.g., on-the-fly). The following section combines... Figure 6 Provides additional details on determining the revised auto-suggested query.

[0081] In various implementations, the lightweight language model 532 is a smaller, simpler, and faster version of the sequence-to-sequence model 514. For example, while the sequence-to-sequence model 514 is an LLM, the lightweight language model 532 is an LSTM or another type of language machine learning model. In multiple implementations, the lightweight language model 532 is a classifier-type model that operates within specified low-latency parameters to ensure real-time processing (e.g., providing reformulated auto-suggested queries within milliseconds).

[0082] In some implementations, the online framework 504 excludes or omits the query redefinition model 530. For example, instead of determining the redefinition autosuggestion query, the query redefinition model 530 directly provides the corresponding autosuggestion query to the AI ​​chat system. In some cases, if the redefinition autosuggestion query is presented in the redefinition query cache 534, the query redefinition model 530 provides the redefinition autosuggestion query to the AI ​​chat system; otherwise, it provides the corresponding autosuggestion query to the AI ​​chat system.

[0083] Additionally, the online framework 504 includes activating the AI ​​chat system using a redefined autosuggestion query, as shown in action 540. For example, query gateway system 206 provides the AI ​​chat system with the redefined autosuggestion query along with instructions to cause the AI ​​chat system to execute the redefined autosuggestion query.

[0084] As described above, in various implementations, the AI ​​chat system opens a new, separate user interface. For example, an instance of the AI ​​chat system appears in a new browser tab or window. In some cases, the AI ​​chat system starts within another application. Furthermore, in one or more implementations, the AI ​​chat system opens as a background service (e.g., as a new tab without browser focus) and becomes visible when selected by the user. In different implementations, the AI ​​chat system floats on the surface or becomes visible when an AI chat element is selected, thus triggering a transition to the AI ​​chat system.

[0085] As mentioned above, Figure 6 Additional details are provided regarding the determination of the revised auto-suggested query. Specifically, Figure 6 The diagram illustrates a sample flowchart based on some implementations for generating reformulated autosuggested queries from autosuggested queries. As shown, Figure 6 Including a series of 600 actions together Figure 5 This includes query-reformed model 530. Specifically, a series of actions 600 correspond to the components of query-reformed model 530.

[0086] As shown, a series of actions 600 includes the action 602 of receiving an auto-suggestion query. For example, based on detecting that a user has selected an AI chat element, the query gateway system 206 provides a corresponding auto-suggestion query to the query refactoring model 530.

[0087] Furthermore, a series of actions 600 includes action 604 determining whether an autosuggested query is in the redefined query cache 534. For example, query gateway system 206 queries the redefined query cache 534 to determine whether an autosuggested query matches an entry for an autosuggested query in the redefined query cache 534. If a match is found (e.g., a cache hit), query gateway system 206 identifies one or more redefined autosuggested queries mapped to the matching autosuggested query. In effect, query gateway system 206 identifies redefined autosuggested queries as shown in action 606.

[0088] In some implementations, the query redefinition model 530 is a machine learning model that determines a match based on the proximity of the provided autosuggested query to the autosuggested queries in the redefined query cache 534 within the vector space. For example, the query redefinition model 530 generates a feature vector for the provided autosuggested query and determines whether that feature vector maps to a feature vector of a known autosuggested query within the learned vector space within a threshold distance. In some implementations, the query gateway system 206 determines a match based on a threshold number or quantity of matching or overlapping words.

[0089] If no match is found in the redefined query cache 534 (e.g., cache miss), the query gateway system 206 performs action 608, which utilizes the lightweight language model 532 to determine a redefined autosuggested query at runtime. As mentioned, the lightweight language model 532 generates a new redefined autosuggested query for the provided autosuggested query within a low latency threshold (e.g., milliseconds). Additionally, the query gateway system 206 adds the newly determined redefined autosuggested query to the redefined query cache 534, as shown in action 610.

[0090] In some implementations, the query gateway system 206 also provides autosuggested queries to the classifier model mentioned above to generate reformulated autosuggested queries offline and provides the reformulated autosuggested queries to the reformulated query cache 534. Thus, the reformulated query cache 534 includes the latest version of the reformulated autosuggested queries mapped to the autosuggested queries.

[0091] In some implementations, lightweight language models (or classifier models) also utilize additional context to determine the reformulated autosuggestion query, such as previous searches, neighboring autosuggestion queries, user information, location information, or other information. For example, lightweight language model 532 uses the autosuggestion query to perform a forward search and access the first three results (e.g., title, header information, page information, and / or metadata) and uses this content to generate a better reformulated autosuggestion query for use in the autosuggestion query.

[0092] As mentioned above, by utilizing the redefined query cache 534 to identify redefined auto-suggested queries, the query gateway system 206 avoids reprocessing the same queries, thereby reducing wasteful processing and improving efficiency. Furthermore, by leveraging an online lightweight language model and an offline classifier model, the query gateway system 206 utilizes resource efficiency to produce highly accurate results without sacrificing latency, thus allowing users to wait for the query gateway system 206 to process their requests.

[0093] Figures 7A to 7B The illustration shows an example graphical user interface based on some implementations of a generative language model that leverages multiple autosuggestion queries to generate narrative-based responses. For ease of explanation, Figures 7A to 7B The client device 400 described above includes a client application 401 and a first user interface 402. For example, the first user interface 402 includes a text input 404, an autosuggestion pane 405, an autosuggestion query 406, and the AI ​​chat element 408 described above.

[0094] Furthermore, the client application 401 includes an auto-suggested query selection element 702 and a selection confirmation element 710. For example, in the illustrated embodiment, the query gateway system 206 allows the user to select multiple auto-suggested queries generated for text input 404. As shown, check tags are used to select and indicate multiple auto-suggested queries in auto-suggested query 406.

[0095] Upon detecting the selection of confirmation element 710, query gateway system 206 can trigger the AI ​​chat system as before. For example, query gateway system 206 provides each of the auto-suggested queries corresponding to the selected auto-suggested query selection element to a query redefinition model, which generates a redefinition of the auto-suggested query. In these implementations, the query redefinition model can process each of the provided auto-suggested queries and generate a verbose, comprehensive, and / or generalized redefinition of the auto-suggested query.

[0096] To illustrate, Figure 7B A first user interface 402 of the client application 401 on the client device 400 is shown. Notably, the first user interface 402 of the AI ​​chat system displays a reformulated autosuggestion query 724 (e.g., a single query), which is longer and more verbose than a single selected autosuggestion query. Additionally, the reformulated autosuggestion query 724 is formulated within a comprehensive context-preserving query, enabling the AI ​​chat system to provide a narrative query response 726 that is highly accurate and helpful to the user.

[0097] Although Figure 7A The diagram shows an auto-suggestion query selection element 702 in the auto-suggestion pane of the first user interface 402, but additional implementations may include additional or different elements. For example, the auto-suggestion pane 405 may include an element that opens search results in a new tab or a different AI chat element that triggers various AI chat services. Another example is an auto-suggestion query that includes elements that perform multiple functions, such as opening search results for the auto-suggestion query in a new tab while simultaneously triggering an AI chat system in another new tab.

[0098] Additionally, this disclosure has described an implementation of a query gateway system 206 that provides a seamless gateway between a query search service and an AI chat system. While the search engine example has been used to illustrate an automated query suggestion system, the query gateway system 206 can also work with other types of automated query suggestion systems. These systems may include content-based sites that provide automated query suggestions when a user searches for articles or other content suggesting products and services, or e-commerce sites. Additionally, the query gateway system 206 can operate anywhere that provides automated query suggestions to the user, such as search text fields, browser bars, operating system search boxes, or multi-function search boxes.

[0099] Now go to Figure 8 The diagram illustrates an example flowchart, based on one or more implementations, including a series of actions for utilizing the query gateway system 206. Specifically, Figure 8 The illustration shows a series of example actions based on one or more implementations of a computer-based method for generating narrative query responses using a generative language model.

[0100] Although Figure 8 The diagram illustrates actions based on one or more implementations; however, alternative implementations may omit, add, reorder, and / or modify any of the actions shown. Furthermore, Figure 8 The actions can be performed as part of a method, such as a computer-implemented method. Alternatively, a non-transitory computer-readable medium can include actions that cause a computing device to execute when executed by a processing system including a processor. Figure 8 The instructions are the actions of the system. In a further implementation, the system can execute (e.g., a processing system with a processor can make the instructions execute). Figure 8 The action.

[0101] As shown, a series of actions 800 includes an action 810 of generating an eligibility cache for autosuggested queries. For example, action 810 involves generating a generative language model eligibility cache for autosuggested queries. In various implementations, action 810 includes generating a generative language model eligibility cache for autosuggested queries by classifying the autosuggested queries according to a generative language model.

[0102] In some implementations, action 810 includes: generating a generative language model eligibility cache for auto-suggested queries by classifying auto-suggested queries according to a generative language model; determining that auto-suggested queries are eligible for use in a generative language model by identifying auto-suggested queries in the generative language model eligibility cache; and providing generative language model elements within a first user interface for display adjacent to the auto-suggested queries based on the determination that the auto-suggested queries are eligible for use in a generative language model.

[0103] As further illustrated, a series of actions 800 includes an action 820 of determining an autosuggestion query for the text input. For example, in an example implementation, action 820 involves determining an autosuggestion query from an autosuggestion query set in response to receiving text input corresponding to a search query. In one or more implementations, action 820 includes determining an autosuggestion query from an autosuggestion query set in response to receiving text input corresponding to a search query.

[0104] In various implementations, action 820 includes: receiving text input corresponding to a search query from a client device. In various implementations, action 820 includes: determining an auto-suggestion query from an auto-suggestion query set stored in an auto-suggestion query database based on the text input. In one or more implementations, action 820 includes a generative language model qualification cache for generating auto-suggestion queries based on query logs and a classifier model from previous text input.

[0105] As further illustrated, a series of actions 800 includes action 830 of providing GLM elements immediately adjacent to the autosuggested query based on determining that the autosuggested query is eligible for use in a generative language model (GLM). For example, in an example implementation, action 830 involves providing generative language model elements within a first user interface for display immediately adjacent to the autosuggested query, based on determining that the autosuggested query is eligible for use in a generative language model using a generative language model eligibility cache.

[0106] In one or more implementations, action 830 includes providing a generative language model element for display immediately adjacent to an autosuggested query within a first user interface. In some implementations, action 810 includes providing a generative language model element immediately adjacent to an autosuggested query within the first user interface, based on determining that the autosuggested query is eligible using a generative language model eligibility cache. In various implementations, action 830 includes determining that an autosuggested query is eligible for use with the generative language model by identifying autosuggested queries in the generative language model eligibility cache.

[0107] In some implementations, action 830 includes: determining that the additional autosuggestion query is ineligible for use in the generative language model. Based on the ineligibility of the additional autosuggestion query for use in the generative language model, action 830 further includes: determining that no generative language model element is provided for display immediately adjacent to the additional autosuggestion query within the first user interface. In various implementations, action 830 includes: displaying a generative language model element with an autosuggestion query in an autosuggestion user interface pane, wherein the autosuggestion user interface pane includes a second autosuggestion query displayed using a second generative language model element and a third autosuggestion query displayed in the absence of any generative language model element.

[0108] As further illustrated, a series of actions 800 includes action 840 of generating a reformulated autosuggestion query from the autosuggestion query. For example, in the example implementation, action 840 involves generating a reformulated autosuggestion query from the autosuggestion query in response to detecting the selection of a generative language model element.

[0109] In one or more implementations, action 840 includes: in response to detecting the selection of a generative language model element in a first user interface, generating a revised auto-suggested query from the auto-suggested query using a revision model that includes a lightweight language model and a revised query cache. In some implementations, action 840 includes: in response to detecting the selection of a generative language model element in a first user interface, generating a revised auto-suggested query from the auto-suggested query using a revision model and providing the revised auto-suggested query to the generative language model as an auto-suggested query.

[0110] In various implementations, action 840 includes: generating a revised autosuggested query from an autosuggested query by determining that the autosuggested query is in the revised query cache and identifying the revised autosuggested query in the revised query cache associated with the autosuggested query. In some implementations, action 840 includes: generating a revised autosuggested query for the revised query cache from a previous autosuggested query using a sequence-to-sequence machine learning model. According to some implementations, a revised autosuggested query is generated from an autosuggested query by determining that the autosuggested query is not in the revised query cache, using a lightweight language model to determine the revised autosuggested query at runtime, and adding the revised autosuggested query to the revised query cache used for the autosuggested query.

[0111] In some implementations, action 840 includes: generating a redefined auto-suggested query from the auto-suggested query by determining that the auto-suggested query is in the redefined query cache and identifying the redefined auto-suggested query in the redefined query cache associated with the auto-suggested query. In various implementations, the redefined auto-suggested query is more verbose than the auto-suggested query.

[0112] As further illustrated, a series of actions 800 includes action 850 of providing a reformulated autosuggestion query to the GLM. For example, in the example implementation, action 850 involves providing a reformulated autosuggestion query to the generative language model for display in a second user interface separate from the first user interface.

[0113] In one or more implementations, action 850 includes: in response to detecting the selection of a generative language model element, providing an autosuggested query to the generative language model for display in a second user interface separate from the first user interface. In some implementations, action 850 includes: providing a reformulated autosuggested query to the generative language model along with a descriptive query result of the reformulated autosuggested query generated by the generative language model for display in a second user interface separate from the first user interface.

[0114] In some implementations, providing a reformulated autosuggestion query to the generative language model causes the generative language model to automatically generate a narrative query response to the reformulated autosuggestion query. In various implementations, action 850 includes: opening a new browser tab in a browser on the user's client device, which displays a second user interface of the generative language model, wherein the second user interface includes the reformulated autosuggestion query and the narrative query response.

[0115] In various implementations, action 850 includes: detecting additional selections of additional generative language model elements immediately adjacent to the additional autosuggestion query displayed for text input and causing an additional new browser tab to open in a browser on the user's client device, the browser displaying additional instances of the generative language model including the additional autosuggestion query and the additional narrative query response to the additional autosuggestion query.

[0116] In some implementations, the series of actions 800 includes additional actions. For example, the series of actions 800 includes the following actions: detecting the selection of multiple generative language model elements corresponding to multiple auto-suggestion queries provided in response to text input, detecting additional selections of combined generative language model elements, generating a revised auto-suggestion query from the multiple auto-suggestion queries, and providing the revised auto-suggestion query to the generative language model.

[0117] Figure 9 The illustration shows certain components that may be included within a computer system 900. The computer system 900 can be used to implement the various computing devices, components, and systems described herein (e.g., by executing computer-implemented instructions). As used herein, "computing device" refers to an electronic component that performs a set of operations based on a set of programmed instructions. Computing devices include the group consisting of electronic components, client devices, server devices, etc.

[0118] In various implementations, computer system 900 refers to one or more of the client devices, server devices, or other computing devices described above. For example, computer system 900 can refer to various types of network devices capable of accessing data on a network, a cloud computing system, or another system. For example, client device can refer to mobile devices such as mobile phones, smartphones, personal digital assistants (PDAs), tablet computers, laptop computers, or wearable computing devices (e.g., headsets or smartwatches). Client device can also refer to non-mobile devices such as desktop computers, server nodes (e.g., from another cloud computing system), or other non-portable devices.

[0119] Computer system 900 includes a processing system, which includes processor 901. Processor 901 can be a general-purpose single-chip or multi-chip microprocessor (e.g., an Advanced Reduced Instruction Set Computer (RISC) machine (ARM)), a special-purpose microprocessor (e.g., a Digital Signal Processor (DSP)), a microcontroller, a programmable gate array, etc. Processor 901 can be referred to as a Central Processing Unit (CPU) and enables the execution of instructions implemented by the computer. Although the processor 901 shown is only... Figure 9 The computer system 900 uses a single processor, but in alternative configurations, a combination of processors (e.g., ARM and DSP) can be used.

[0120] The computer system 900 also includes a memory 903 that is in electronic communication with the processor 901. The memory 903 can be any electronic component capable of storing electronic information. For example, the memory 903 can be implemented as random access memory (RAM), read-only memory (ROM), disk storage media, optical storage media, flash memory devices in RAM, onboard memory included in the processor, erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, etc., including combinations thereof.

[0121] Instruction 905 and data 907 may be stored in memory 903. Instruction 905 may be executed by processor 901 to implement some or all of the functions disclosed herein. Execution of instruction 905 may involve using data 907 stored in memory 903. Any example of the various examples of modules and components described herein may be implemented in part or in whole as instruction 905 stored in memory 903 and executed by processor 901. Any example of the various examples of data described herein may be within data 907, which is stored in memory 903 and used during the execution of instruction 905 by processor 901.

[0122] The computer system 900 may also include one or more communication interfaces 909 for communicating with other electronic devices. The one or more communication interfaces 909 may be based on wired communication technology, wireless communication technology, or both. Some examples of the one or more communication interfaces 909 include Universal Serial Bus (USB), Ethernet adapters, wireless adapters operating according to the Institute of Electrical and Electronics Engineers (IEEE) 902.11 wireless communication protocol, Bluetooth® wireless communication adapters, and infrared (IR) communication ports.

[0123] Computer system 900 may also include one or more input devices 911 and one or more output devices 913. Some examples of the one or more input devices 911 include a keyboard, mouse, microphone, remote control device, button, joystick, trackball, touchpad, and light pen. Some examples of the one or more output devices 913 include speakers and printers. A particular type of output device typically included in computer system 900 is a display device 915. The display device 915 used with the implementation disclosed herein can utilize any suitable image projection technology, such as liquid crystal display (LCD), light-emitting diode (LED), gas plasma, electroluminescence, etc. A display controller 917 may also be provided for converting data 907 stored in memory 903 into text, graphics, and / or moving images (where appropriate) displayed on display device 915.

[0124] The various components of the computer system 900 can be coupled together via one or more buses, which may include power buses, control signal buses, status signal buses, data buses, etc. For clarity, in... Figure 9 The diagram of each bus is shown as Bus System 919.

[0125] This disclosure describes a query gateway system in a network architecture. In this disclosure, "network" refers to one or more data links that enable the electronic data transfer between computer systems, modules, and other electronic devices. A network can include public networks such as the Internet as well as private networks. When information is transmitted or provided through a network or another communication connection (hardwired, wireless, or both), the computer correctly regards that connection as a transmission medium. The transmission medium can include a network and / or data link carrying the required program code in the form of computer-executable instructions or data structures, which can be accessed by general-purpose or special-purpose computers.

[0126] Furthermore, the networks described herein can refer to networks or combinations of networks (e.g., the Internet, corporate intranet, virtual private network (VPN), local area network (LAN), wireless local area network (WLAN), cellular network, wide area network (WAN), metropolitan area network (MAN), or a combination of two or more such networks), through which one or more computing devices can access the various systems described in this disclosure. In practice, the networks described herein can include one or more networks using one or more communication platforms or technologies for transmitting data. For example, a network can include the Internet or other data links that enable the transmission of electronic data between corresponding client devices and components (e.g., server devices and / or virtual machines on them) of a cloud computing system.

[0127] Furthermore, upon arrival at various computer system components, program code in the form of computer-executable instructions or data structures can be automatically transferred from the transmission medium to a non-transitory computer-readable storage medium (device) and vice versa. For example, computer-executable instructions or data structures received via a network or data link can be cached in random access memory (RAM) within a network interface module (NIC) and then ultimately transferred to the computer system RAM and / or a less volatile computer storage medium (device) at the computer system. Therefore, it should be understood that non-transitory computer-readable storage media (devices) can be included in computer system components that also (or even primarily) utilize the transmission medium.

[0128] Computer-executable instructions include instructions and data that, when executed by a processor, cause a general-purpose computer, a special-purpose computer, or a special-purpose processing device to perform a particular function or group of functions. In some implementations, the computer-executable and / or computer-implemented instructions are executed by a general-purpose computer to transform the general-purpose computer into a special-purpose computer that implements the elements of this disclosure. Computer-executable instructions may include, for example, binary code, intermediate format instructions such as assembly language, or even source code. Although the subject matter has been described in language specific to structural features and / or methodological actions, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the features or actions described above. Rather, the described features and actions are disclosed as exemplary forms of implementing the claims.

[0129] Those skilled in the art will appreciate that this disclosure can be practiced in networked computing environments with various types of computer system configurations, including personal computers, desktop computers, laptop computers, message processors, handheld devices, multiprocessor systems, microprocessor-based or programmable consumer electronics, network PCs, minicomputers, mainframes, mobile phones, PDAs, tablets, pagers, routers, switches, etc. This disclosure can also be practiced in distributed system environments, where tasks are performed on both local and remote computer systems linked by a network (via a hardwired data link, a wireless data link, or a combination of hardwired and wireless data links). In a distributed system environment, program modules can reside on both local and remote memory storage devices.

[0130] The techniques described herein can be implemented in hardware, software, firmware, or any combination thereof, unless explicitly described as being implemented in a particular manner. Any features described as modules, components, etc., can also be implemented together in an integrated logic device or separately as discrete but interoperable logic devices. If implemented in software, the techniques can be implemented at least in part by a non-transitory processor-readable storage medium, which includes instructions that, when executed by at least one processor, perform one or more of the methods described herein (including computer-implemented methods). Instructions can be organized into routines, programs, objects, components, data structures, etc., which can perform specific tasks and / or implement specific data types, and can be combined or distributed as desired in various implementations.

[0131] Computer-readable media can be any available medium that can be accessed by a general-purpose or special-purpose computer system. A computer-readable medium storing computer-executable instructions is a non-transitory computer-readable storage medium (device). A computer-readable medium carrying computer-executable instructions is a transmission medium. Therefore, by way of example, implementations of this disclosure may include at least two distinct types of computer-readable media: non-transitory computer-readable storage media (devices) and transmission media.

[0132] As used herein, a non-transitory computer-readable storage medium (device) may include RAM, ROM, EEPROM, CD-ROM, solid-state drive (SSD) (e.g., RAM-based), flash memory, phase-change memory (PCM), other types of memory, other optical disk storage devices, magnetic disk storage devices or other magnetic storage devices, or any other medium that may be used to store desired program code in the form of computer-executable instructions or data structures and that may be accessed by a general-purpose or special-purpose computer.

[0133] Without departing from the scope of the claims, the steps and / or actions of the methods described herein may be interchangeable. In other words, unless proper operation of the described methods requires a specific order of steps or actions, the order and / or use of specific steps and / or actions may be modified without departing from the scope of the claims.

[0134] The term "determine" encompasses a variety of actions, and therefore "determine" can include calculation, operation, processing, derivation, investigation, searching (e.g., looking in a table, database, or other data structure), ascertainment, etc. Furthermore, "determine" can include receiving (e.g., receiving information), accessing (e.g., accessing data in memory), etc. Additionally, "determine" can include parsing, selecting, picking, building, etc.

[0135] The terms “comprising,” “including,” and “having” are intended to be inclusive and mean that additional elements may exist in addition to those listed. Additionally, it should be understood that references to “one implementation” or “multiple implementations” in this disclosure are not intended to be construed as excluding the existence of additional implementations that also incorporate the described features. For example, where compatible, any element or feature described with respect to an implementation herein may be combinable with any element or feature of any other implementation described herein.

[0136] This disclosure may be embodied in other specific forms without departing from its spirit or characteristics. The described implementations are to be considered illustrative rather than restrictive. The scope of this disclosure is indicated by the appended claims rather than by the foregoing description. Changes in the meaning and scope of equivalents of the claims will be included within their scope.

Claims

1. A computer-implemented method for generating narrative query responses with a generative language model, comprising: generating a generative language model eligibility cache (120) of auto-suggested queries; in response to receiving text input (110) corresponding to a search query, determining an auto-suggested query (124) from a set of auto-suggested queries (112); based on determining that the auto-suggested query (124) is eligible for a generative language model (130) using the generative language model eligibility cache (120), providing a generative language model element (122) for display proximate to the auto-suggested query (124) within a first user interface (116); in response to detecting a selection of the generative language model element (122), generating a reformulated auto-suggested query (126) from the auto-suggested query (124); and providing the reformulated auto-suggested query (124) to the generative language model (130) for display in a second user interface (132) separate from the first user interface (116).

2. The computer-implemented method of claim 1, further comprising: the generative language model eligibility cache generates auto-suggested queries based on a query log of previous text inputs and a classifier model.

3. The computer-implemented method of claim 1, further comprising: determining that the auto-suggested query is eligible for the generative language model by identifying the auto-suggested query in the generative language model eligibility cache.

4. The computer-implemented method of claim 1, further comprising: determining that an additional auto-suggested query is not eligible for the generative language model; and based on the additional auto-suggested query not being eligible for the generative language model, determining not to provide any generative language model element for display proximate to the additional auto-suggested query within the first user interface.

5. The computer-implemented method of claim 1, further comprising: displaying the generative language model element with the auto-suggested query in an auto-suggested user interface pane, wherein the auto-suggested user interface pane includes a second auto-suggested query displayed with a second generative language model element and a third auto-suggested query not displayed with any generative language model element.

6. The computer-implemented method of claim 1, further comprising: generating the reformulated auto-suggested query from the auto-suggested query by: determining that the auto-suggested query is in a reformulated query cache; and identifying the reformulated auto-suggested query in the reformulated query cache associated with the auto-suggested query.

7. The computer-implemented method of claim 6, further comprising: generating reformulated auto-suggested queries for the reformulated query cache from previous auto-suggested queries with a sequence-to-sequence machine learning model.

8. The computer-implemented method of claim 1, further comprising: generating the reformulated auto-suggested query from the auto-suggested query by: determining that the auto-suggested query is not in a reformulated query cache; determining the reformulated auto-suggested query on the fly with a lightweight language model; and adding the reformulated auto-suggested query to the reformulated query cache for the auto-suggested query.

9. The computer-implemented method of claim 1, wherein the reformulated autosuggest query is more verbose than the autosuggest query.

10. The computer-implemented method of claim 1, wherein providing the reformulated autosuggest query to the generative language model causes the generative language model to automatically generate a narrative query response to the reformulated autosuggest query.

11. The computer-implemented method of claim 10, further comprising: causing a new browser tab to open in a browser on the user's client device, the new browser tab showing the second user interface of the generative language model, wherein the second user interface includes the reformulated autosuggest query and the narrative query response.

12. The computer-implemented method of claim 11, further comprising: detecting an additional selection of an additional generative language model element displayed proximate to the additional autosuggest query for the text input; and causing an additional new browser tab to open in the browser on the user's client device, the additional new browser tab showing an additional instance of the generative language model, the additional instance including the additional autosuggest query and an additional narrative query response responsive to the additional autosuggest query.

13. A computer-implemented method for generating narrative query responses with a generative language model, comprising: receiving a text input (110) corresponding to a search query from a client device (230); determining an autosuggest query (124) from a set of autosuggest queries (112) based on the text input (110), the set of autosuggest queries (112) stored in an autosuggest query cache; providing a generative language model element (122) for display proximate to the autosuggest query (124) within a first user interface (116); responsive to detecting a selection of the generative language model element (122) in the first user interface (116), generating a reformulated autosuggest query (126) from the autosuggest query (124) with a reformulation model (530) having a light-weight language model (532) and a reformulated query cache (534); and providing the reformulated autosuggest query (126) to a generative language model (130) for display in a second user interface (132) separate from the first user interface (116) along with narrative query results for the reformulated autosuggest query (124) generated by the generative language model (130).

14. The computer-implemented method of claim 13, further comprising: generating a generative language model eligibility cache of autosuggest queries by classifying the autosuggest queries according to the generative language model; determining that the autosuggest query is eligible for the generative language model by identifying the autosuggest query in the generative language model eligibility cache; and based on determining that the autosuggest query qualifies for the generative language model, providing the generative language model element for display proximate to the autosuggest query within the first user interface.

15. The computer-implemented method of claim 13, further comprising: detecting a selection of a plurality of generative language model elements corresponding to a plurality of autosuggest queries provided in response to the text input; detecting an additional selection of a combined generative language model element; generating the reformulated autosuggest query from the plurality of autosuggest queries; and providing the reformulated autosuggest query to the generative language model.

16. The computer-implemented method of claim 13, further comprising: generating the reformulated autosuggest query from the autosuggest query by: determining that the autosuggest query is in the reformulated query cache; and identifying the reformulated autosuggest query in the reformulated query cache associated with the autosuggest query.

17. The computer-implemented method of claim 16, further comprising: generating reformulated autosuggest queries for the reformulated query cache from previous autosuggest queries with a sequence-to-sequence machine learning model.

18. The computer-implemented method of claim 13, further comprising: generating the reformulated autosuggest query from the autosuggest query by: determining that the autosuggest query is not in the reformulated query cache; utilizing the lightweight language model to determine the reformulated autosuggest query on the fly; and adding the reformulated autosuggest query to the reformulated query cache for the autosuggest query.

19. A system for generating narrative query responses with a generative language model, comprising: a processor (901); and a computer memory (903) comprising instructions (905) that, when executed by the processor (901), cause the system to perform operations comprising: generating a generative language model eligibility cache (120) of autosuggest queries (114) by classifying the autosuggest queries (114) according to a generative language model (130); in response to receiving a text input (110) corresponding to a search query, determining an autosuggest query (124) from a set of autosuggest queries (112); based on determining that the autosuggest query (124) qualifies for a generative language model (130) using the generative language model eligibility cache (120), providing a generative language model element (122) for display proximate to the autosuggest query (124) within a first user interface (116); and in response to detecting a selection of the generative language model element (122), providing the autosuggest query (124) to the generative language model (130) for display in a second user interface (132) separate from the first user interface (116).

20. The system of claim 19, further comprising instructions that, when executed by the processor, cause the system to perform operations comprising: In response to detecting the selection of the generative language model element in the first user interface, generating a reformulated autosuggest query from the autosuggest query utilizing a reformulation model; and providing the reformulated autosuggest query to the generative language model as the autosuggest query.

Images