Information generation method and device

Abstract

The invention discloses an information generation method and device. Belongs to the technical field of artificial intelligence. The method comprises the following steps: receiving question information of a user; scene description information and user memory information associated with the question information are obtained, the user memory information comprises event description information and event sequence description information, the event description information is used for describing events associated with the question information, and the event sequence description information is used for describing event sequences associated with the events; and inputting the question information, the scene description information and the user memory information into the large language model to obtain answer information which is generated by the large language model and aims at the question information.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of artificial intelligence technology, and specifically to an information generation method and device thereof. Background Art

[0002] Large Language Model (LLM) refers to a language model with a large number of parameters and capable of processing large-scale language data. At present, artificial intelligence question-answering assistants based on large language models can efficiently answer questions raised by users and help users obtain information.

[0003] In the prior art, usually only the user's question information and the historical conversation content are input into the large language model to obtain the answer information output by the large language model. However, the user's conversation content only occupies a small part of the user's personal memory. It is difficult to obtain accurate and comprehensive answer information by only using this information to guide the large language model to answer questions. Summary of the invention

[0004] The purpose of the embodiments of the present application is to provide an information generation method and device thereof, which can improve the comprehensiveness and accuracy of answer information generated by a large language model in a question-answering scenario.

[0005] In a first aspect, an embodiment of the present application provides an information generation method, the method comprising: receiving a user's question information; obtaining scene description information of a historical scene associated with the question information and user memory information associated with the question information, the user memory information comprising event description information and event sequence description information, the event description information being used to describe an event associated with the question information, and the event sequence description information being used to describe an event sequence associated with the event; inputting the question information, the scene description information and the user memory information into a large language model to obtain answer information for the question information generated by the large language model.

[0006] In a second aspect, an embodiment of the present application provides an information generating device, which includes: a receiving unit, used to receive a user's question information; an acquiring unit, used to acquire scene description information of historical scenes associated with the question information and user memory information associated with the question information, the user memory information including event description information and event sequence description information, the event description information being used to describe the event associated with the question information, and the event sequence description information being used to describe the event sequence associated with the event; a generating unit, used to input the question information, the scene description information and the user memory information into a large language model to obtain answer information for the question information generated by the large language model.

[0007] In a third aspect, an embodiment of the present application provides an electronic device, which includes a processor and a memory, wherein the memory stores programs or instructions that can be run on the processor, and when the program or instructions are executed by the processor, the steps of the method described in the first aspect are implemented.

[0008] In a fourth aspect, an embodiment of the present application provides a readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method described in the first aspect above are implemented.

[0009] In a fifth aspect, an embodiment of the present application provides a chip, comprising a processor and a communication interface, wherein the communication interface is coupled to the processor, and the processor is used to run a program or instruction to implement the method described in the first aspect.

[0010] In a sixth aspect, an embodiment of the present application provides a computer program product, which is stored in a storage medium and is executed by at least one processor to implement the method described in the first aspect.

[0011] In an embodiment of the present application, after receiving the user's question information, the scene description information associated with the question information and the user memory information associated with the question information are first obtained, and then the question information, the scene description information and the user memory information are input into the large language model, so as to obtain the answer information for the question information generated by the large language model. Among them, the user memory information includes event description information for describing the event associated with the question information and event sequence description information for describing the event sequence associated with the event. Since the information input into the large language model includes, in addition to the question information, the scene description information associated with the question, the event description information of the event associated with the question, and the event sequence description information of the event sequence associated with the event, the above information can deeply restore the user memory related to the question raised by the user, so the information input into the large language model is more extensive and in-depth, so that the answers to the questions generated by the large language model with the help of the above information have higher accuracy and comprehensiveness. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 is a flow chart of the information generation method provided in an embodiment of the present application;

[0013] Figure 2 It is a schematic diagram of a scenario library of the information generation method provided in an embodiment of the present application;

[0014] Figure 3 is a schematic diagram of a user memory network of the information generation method provided in an embodiment of the present application;

[0015] Figure 4It is a structural schematic diagram of an information generating device provided in an embodiment of the present application;

[0016] Figure 5 is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application;

[0017] Figure 6 It is a schematic diagram of the hardware structure of an electronic device suitable for implementing the embodiments of the present application. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present application to clearly describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments in the present application belong to the scope of protection of this application.

[0019] The terms "first", "second", etc. in the specification and claims of the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described here, and the objects distinguished by "first", "second", etc. are generally of one type, and the number of objects is not limited. For example, the first object can be one or more. In addition, "and / or" in the specification and claims represents at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0020] The information generation method and device provided in the embodiments of the present application are described in detail below through specific embodiments and their application scenarios in conjunction with the accompanying drawings.

[0021] Please refer to Figure 1 , which shows one of the flow charts of the information generation method provided in the embodiment of the present application. The information generation method provided in the embodiment of the present application can be applied to electronic devices. In practice, the above-mentioned electronic devices can be electronic devices such as smart phones, tablet computers, laptops, and wearable devices.

[0022] The process of the information generation method provided in the embodiment of the present application includes the following steps:

[0023] Step 101, receiving user's question information.

[0024] In this embodiment, the question information can be text directly input by the user, or can be text obtained by converting the voice input by the user. As an example, the embodiment of the present application can be applied to a dialogue scenario based on artificial intelligence. The question information can be "What memories does the music festival bring me?", "What food do I like to eat?", etc.

[0025] Step 102, obtain scene description information of historical scenes associated with the problem information and user memory information associated with the problem information, the user memory information includes event description information and event sequence description information, the event description information is used to describe the event associated with the problem information, and the event sequence description information is used to describe the event sequence associated with the event.

[0026] In this embodiment, the scene description information of the historical scene associated with the question information may be first obtained. The scene description information may be used to describe the scene related to the question information. The scene description information may be information in text form.

[0027] As an example, the question information is “What memories does the music festival bring me?”, then the historical scene related to the question information may be a scene of the user attending a music festival, and the scene description information may be “the user is attending a music festival”.

[0028] As another example, the question information is "What kind of food do I like to eat?", then the historical scene related to the question information may be a scene in which the user has dined out. If the user has dined at a barbecue restaurant and a noodle restaurant, then the scene description information may be "the user dined at a barbecue restaurant, the user dined at a noodle restaurant."

[0029] In this embodiment, user memory information associated with the question information may also be obtained. The user memory information may include event description information and event sequence description information. The event description information may be used to describe an event associated with the question information. The event may be one or more historical events. The event sequence description information may be used to describe an event sequence associated with the event. The event sequence may include the event. Both the event description information and the event sequence description information may be information in text form, and their formats may be set.

[0030] In some scenarios, events in an event sequence have temporal correlation. Temporal correlation between events may mean that events occur in sequence and the occurrence of the latter event depends on the previous event. At this time, the event sequence description information can be used to describe the order of events in the event sequence. As an example, the question information is "What food do I like to eat?", then the user memory information associated with the question information may include event description information of the event that the user once participated in a music festival, for example, {"Date":"June 16, 2024 13:30","Location":"Fuzhou Minyue Water Town Festival Square","Title":"Bubble Island Music and Art Festival","Description":"These photos show a music festival called 'Heineken BubblingBoiling Festival', where people enjoy outdoor stage performances in the rain, including singers such as XX"}. In addition to the event, the event sequence related to the event may also include events such as departing from the permanent residence, taking the high-speed rail to the music festival venue, tasting local food, and returning to the permanent residence. Accordingly, an example of event sequence description information is as follows: {The time range of this trip to Fuzhou City is from June 15 to June 17, 2024. Departing from Hangzhou, taking the high-speed rail to Fuzhou, tasting the food of XX barbecue restaurant, participating in the Paopao Island Music and Art Festival, people wearing wristbands enjoyed the performances of singers such as XX outdoors in the rain, and enjoying the box lunch and shows on the train on the way back.}.

[0031] In some scenarios, the events in the event sequence are of the same category and have no time correlation. At this time, the event sequence description information can be a user description of the user preferences determined based on the events in the event sequence. As an example, the question information is "What kind of food do I like to eat?", then the user memory information associated with the question information may include event description information of events in which the user has dined out, for example, it may include: {"Date":"June 1, 2024 12:59:29","Location":"XX Noodle House (Cangqian Tang Road Store)","Title":"Dining at XX Noodle House (Cangqian Tang Road Store) at noon","Description":"At noon on June 1, 2024, I enjoyed a delicious lunch at XX Noodle House (Cangqian Tang Road Store), including a bowl of noodles, a drink and a bun."}, {"Date":"June 15, 2024 23:59:0 6","Location":"XX Barbecue (Jinshan Branch)","Title":"Dining at XX Barbecue (Jinshan Branch) in the evening","Description":"On the evening of June 15, 2024, a group of people enjoyed barbecue at XX Barbecue, and the table was full of side dishes and condiments. "}, {"Date":"May 16, 2024 17:35:06","Location":"XXX Barbecue (Pingfeng Street Branch)","Title":"Dining at XXX Barbecue (Pingfeng Street Branch) in the evening","Description":"On the evening of May 16, 2024, my friends and I ate barbecue at XXX Barbecue, and there were seafood cakes and cakes on the table. "}. The above-mentioned events in which the user has dined out may constitute an event sequence. The event sequence description information may include user preference information determined based on the above-mentioned event sequence, for example, {in terms of food, the user prefers barbecue}.

[0032] Step 103: input the question information, scene description information and user memory information into the large language model to obtain answer information for the question information generated by the large language model.

[0033] In this embodiment, the large language model refers to a neural network model containing ultra-large-scale parameters. The large language model has a huge scale, multi-task learning capabilities, powerful computing resources and rich training data. The large language model is widely used in the field of natural language processing and has the ability to perform well in various natural language processing tasks, such as question-answering tasks, text classification tasks, sentiment analysis tasks, summary generation tasks, translation tasks, etc. The large language model can be used in multiple application fields such as automatic writing, chatbots, virtual assistants, voice assistants, and automatic translation. Since the large language model has a more outstanding performance in natural language processing tasks, the generation of answers to questions with the help of a large language model can more fully and accurately understand the semantics of the input information compared to other models, making the output answer more accurate.

[0034] In this embodiment, the question information, the acquired scene description information and the user memory information can be first substituted into the instruction template to obtain the input instruction. The input instruction refers to the instruction input into the large language model. The input instruction is used to require the model to output the expected answer in a specified format. Then, the input instruction is input into the large language model to obtain the answer information generated by the large language model for the question information.

[0035] As an example, the question information is "What memories does the music festival bring me?", the scene description information is "The user is attending a music festival", and the user memory information includes {"Date":"June 16, 2024 13:30","Location":"Fuzhou Minyue Water Town Festival Square","Title":"Bubble Island Music and Art Festival","Description":"These photos show a music festival called 'Heineken Bubbling Boiling Festival', where people enjoy outdoor stage performances in the rain, including singers such as XX"}{The time range of this Fuzhou trip is from June 15 to June 17, 2024. Departing from Hangzhou, take the high-speed rail to Fuzhou, taste the food of XX barbecue restaurant, participate in the Bubble Island Music and Art Festival, people wear bracelets to enjoy the performances of singers such as XX outdoors in the rain, and enjoy the box lunch and show on the train on the way back.}, then the input command can be referred to as follows:

[0036] “You are a personalized Q&A assistant in the user’s phone. You will help answer the user’s questions based on the relevant content retrieved from the user’s phone.

[0037] If you don’t know how to answer this question or the relevant content you retrieved is not helpful in answering this question, you can answer “I don’t know”.

[0038] Your answer should be as detailed as possible.

[0039]

User's Question

[0040]

Scene related to the question

[0041]

Retrieved user memories

[0042] {"Date": "June 16, 2024, 13:30", "Location": "Festival Square, Minyue Water Town, Fuzhou", "Title": "Bubble Island Music & Arts Festival", "Description": "These photos show the scene of a music festival named 'Heineken Bubbling Boiling Festival'. People enjoyed the outdoor stage performance in the rain. The scene included singers such as XX"}

[0043] {The time range of this trip to Fuzhou was from June 15 to June 17, 2024. Starting from Hangzhou, taking the high - speed train to Fuzhou, tasting the food at XX barbecue restaurant, participating in the Bubble Island Music & Arts Festival. People wore bracelets and enjoyed the performances of singers such as XX outdoors in the rain, and enjoyed the boxed meals and programs on the train during the return journey.}

[0044] Your answer: ”.

[0045] In the above example, after inputting the above content into the large - language model, the answer information output by the large - language model can be: “You attended the 'Bubble Island Music & Arts Festival' held at the Festival Square, Minyue Water Town, Fuzhou on June 16, 2024. There, you and others wore admission bracelets and enjoyed the live performances of singers such as XX in the rain.”.

[0046] As another example, the question information is "What kind of food do I like to eat?", the scene description information is "The user eats at a barbecue restaurant, and the user eats at a noodle restaurant", and the user memory information includes {In terms of food, the user prefers barbecue}, {"Date":"June 1, 2024 12:59:29","Location":"XX Noodle House (Cangqian Tang Road Store)","Title":"Dining at XX Noodle House (Cangqian Tang Road Store) at noon","Description":"At noon on June 1, 2024, I enjoyed a delicious lunch at XX Noodle House (Cangqian Tang Road Store), including a bowl of noodles, a drink and a bun."}, {"Date":"June 15, 2024 23:59:06", "Location":"XX Barbecue (Jinshan Branch)","Title":"Dining at XX Barbecue (Jinshan Branch) in the evening","Description":"On the evening of June 15, 2024, a group of people enjoyed barbecue at XX Barbecue Restaurant, and the table was full of side dishes and condiments."}、{"Date":"May 16, 2024 17:35:06","Location":"XXX Barbecue (Pingfeng Street Branch)","Title":"Dining at XXX Barbecue (Pingfeng Street Branch) in the evening","Description":"On the evening of May 16, 2024, my friends and I ate barbecue at XXX Barbecue Restaurant, and there were seafood cakes and cakes on the table."}, then the input command can be referred to as follows:

[0047] “You are a personalized Q&A assistant in the user’s phone. You will help answer the user’s questions based on the relevant content retrieved from the user’s phone.

[0048] If you don’t know how to answer this question or the relevant content you retrieved is not helpful in answering this question, you can answer “I don’t know”.

[0049] Your answer should be as detailed as possible.

[0050]

User's question

[0051] [Scenarios related to the question]: The user is dining at a barbecue restaurant, and the user is dining at a noodle restaurant.

[0052]

Retrieved user memory

[0053] {In terms of food, users prefer barbecue}

[0054] {"Date":"June 1, 2024 12:59:29","Location":"XX Noodle House (Cangqiantang Road Store)","Title":"Having lunch at XX Noodle House (Cangqiantang Road Store) at noon","Description":"At noon on June 1, 2024, I had a delicious lunch at XX Noodle House (Cangqiantang Road Store), including a bowl of noodles, a cup of drink and a bun."}

[0055] {"Date":"June 15, 2024 23:59:06","Location":"XX Barbecue (Jinshan Store)","Title":"Having dinner at XX Barbecue (Jinshan Store) in the evening","Description":"In the evening of June 15, 2024, a group of people had barbecue at XX Barbecue Store, and the table was full of side dishes and seasonings."}

[0056] {"Date":"May 16, 2024 17:35:06","Location":"XXX Barbecue (Pingfeng Street Store)","Title":"Having dinner at XXX Barbecue (Pingfeng Street Store) in the evening","Description":"In the evening of May 16, 2024, I had barbecue with my friends at XXX Barbecue Store, and there were seafood pancakes and cakes on the table."}

[0057] Your answer:".

[0058] In the above example, after inputting the above content into the large language model, the answer information output by the large language model can be: "You like to eat barbecue the most, such as having barbecue at XX Barbecue Store, XXX Barbecue Store, etc. In addition, you also seem to like eating noodles, such as the noodles and buns you had at XX Noodle House."

[0059] For the method provided in the above embodiments of the present application, after receiving the question information of the user, first obtain the scenario description information associated with the question information and the user memory information associated with the question information, and then input the question information, the scenario description information and the user memory information into the large language model, so as to obtain the answer information generated by the large language model for the question information. Since the information input into the large language model includes, in addition to the question information, the scenario description information associated with the question, the event description information of the event associated with the question, and the event sequence description information of the event sequence associated with the event, the above scenario description information, event description information and event sequence description information can deeply restore the user memory related to the question raised by the user. Therefore, the information input into the large language model is more extensive and in-depth, so that the answer to the question generated by the large language model with the help of the above information has higher accuracy and comprehensiveness.

[0060] In some optional embodiments, in the above step 102, obtaining the scene description information of the historical scene associated with the question information may include: obtaining the scene description information of the historical scene associated with the question information from the scene library. Here, the scene description information may be retrieved from the scene library through a variety of retrieval methods such as similarity retrieval and full-text retrieval to obtain the scene description information of the historical scene associated with the question information.

[0061] The scene library can be generated in advance through the following sub-steps:

[0062] Sub-step S11, obtaining basic data of the user in the statistical time period. The statistical time period can be pre-set as needed, for example, it can be set to the most recent year, the most recent month, etc. The basic data includes at least one of the following: environmental data, user data, and device data. Among them, environmental data may include but is not limited to weather information, time, location, etc. User data may include but is not limited to user attribute information, status information, and behavioral data generated by using the device. The above attribute information may include but is not limited to gender, age, etc. The above status information may include but is not limited to heart rate, number of steps, etc. The above behavioral data may include but is not limited to the content clicked by the user, the content browsed, the stored images, etc. The device data may include but is not limited to the status information of the electronic device, system events, and posture information. The status of the electronic device may include but is not limited to the network connection status, charging status, etc.

[0063] Sub-step S12, based on the basic data, generates scene description information for multiple historical scenes. The historical scenes can be determined based on events, locations, behavior data, etc. in the basic data. The scene description information of each historical scene can be used to describe the historical scene. For example, for the arrival in Fuzhou scene, the scene description information can be "the user arrives in Fuzhou". For the on-the-go scene, the scene description information can be "the user is on the road".

[0064] Sub-step S13, storing the generated scene description information according to the set rules to obtain a scene library. Here, scene description information of the same category can be stored in one sub-library.

[0065] By pre-creating a scenario library and retrieving scenario description information of historical scenarios associated with the question information from the scenario library, accurate scenario description information of historical scenarios associated with the question information can be obtained conveniently and quickly.

[0066] In some optional embodiments, in the above sub-step S13, the category of each historical scene in the multiple historical scenes may be determined first. Then, the scene description information of the historical scenes belonging to the same category in the multiple historical scenes is stored in the same scene sub-library. Finally, a scene library is generated based on the scene sub-libraries corresponding to each category. As an example, see Figure 2, the categories of historical scenes include but are not limited to at least one of the following: life, work, travel, social, health, entertainment, etc. Among them, life scenes may further include but are not limited to scenes before important dates, appointment registration scenes, scenes after ordering takeout, etc. Work scenes may further include but are not limited to scenes in the company, meeting scenes, overtime scenes, etc. Travel scenes may further include but are not limited to scenes of booking air tickets, arriving at scenic spots, scenes after ordering takeout, etc. The remaining scenes will not be described one by one. By storing the scene description information of historical scenes belonging to the same category in multiple historical scenes into the same scene sub-library, scene matching can be performed conveniently and quickly, so as to efficiently and accurately obtain the scene description information of historical scenes associated with the problem information.

[0067] In some optional embodiments, in the above step 102, obtaining the user memory information associated with the question information may include: obtaining the user memory information associated with the question information from a user memory information library. Here, the user memory information may be retrieved from the user memory information library through a variety of retrieval methods such as similarity retrieval and full-text retrieval to obtain the user memory information of the historical scene associated with the question information.

[0068] The user memory information database is generated in advance through the following steps:

[0069] Sub-step S21, obtaining basic data of the user in the statistical time period. The basic data includes at least one of the following: environmental data, user data, and device data. This step can refer to the above sub-step S11, and will not be repeated here to avoid repetition.

[0070] Sub-step S22, based on the basic data, generates event description information of multiple events. Among them, the event can be determined based on the event, location, behavior data, etc. in the basic data. The event description information of each event can be used to describe the event. The event description information can be extracted in combination with positioning technology, information extraction technology, entity recognition technology, etc.

[0071] Sub-step S23, using events as nodes of the user memory network, determining the connection mode between the nodes based on the association relationship between multiple events, and generating a user memory network based on the nodes and the connection mode. As an example, see Figure 3 If there is a correlation between two events, the nodes corresponding to the two events in the user memory network can be connected. Conversely, if there is no correlation between the two events, the nodes corresponding to the two events in the user memory network are not connected.

[0072] Sub-step S24, for each event in the multiple events, based on the user memory network, determine the event sequence associated with the event, and based on the event description information of each event in the determined event sequence, generate event sequence description information of the determined event sequence. Here, if there is a time correlation between events, the event sequence description information can be used to describe the sequence of events in the event sequence. If there is a time correlation between two events, the event sequence description information can be used to describe the sequence of events in the event sequence. If the events in the event sequence are of the same category, there is no time correlation. At this time, the event sequence description information can be used by the user to describe the user preferences determined based on the events in the event sequence.

[0073] Sub-step S25, the generated event description information of each event and the event sequence description information of each event sequence are summarized to obtain a user memory information library. Each user can have his own user memory information library.

[0074] By constructing a user memory information database and obtaining user memory information associated with question information from the user memory information database, the user's memory can be stored in a structured form in text. The stored user memory information can be deeply restored to the user's memory, making the information input into the large language model more extensive and in-depth, so that the answers to the questions generated by the large language model with the help of the above information have higher accuracy and comprehensiveness.

[0075] In some optional embodiments, the above sub-step S22 may further include:

[0076] The first step is to extract the key information of multiple events from the basic data. The key information of each event includes at least one of the following: the time of the event, the location of the event, and the name of the event. For example, when a user arrives at a music festival, he or she perceives the time information and longitude and latitude information, and can convert the longitude and latitude information into a semantic location. In addition, entity recognition technology can be used to extract the event entity, that is, the name of the music festival, from the notification SMS text sent from the music festival. The final extracted labels can be seen in the table below:

[0077]

[0078] In the second step, for each of the multiple events, an associated image of the event is obtained, where the associated image is an image stored at the time and location of the event.

[0079] In the third step, the associated image is input into the multimodal large language model to obtain image description information for describing the associated image. Among them, the multimodal large language model (MLLM) can realize the semantic understanding of the image and output the image description information. The image description information user describes the content of the image. It should be noted that the large language model in the above embodiment and the multimodal large language model here can be a model with the same structure or a model with a different structure, which is not specifically limited here. As an example, for images taken by users while attending a music festival, the following image description information can be obtained: "These photos show a music festival called 'HeinekenBubbling Boiling Festival', where people enjoy outdoor stage performances in the rain, including singers such as XX."

[0080] The fourth step is to generate event description information of the event based on the key information and image description information of each event. Here, the key information and image description information of each event can be summarized to obtain the event description information of the event.

[0081] By performing semantic understanding on multivariate data, we can obtain the event time, event location, event name, and image description information of each event, making the summarized event description information richer and more comprehensive, and making the answers to questions generated by the large language model with the help of the above event description information more accurate and comprehensive.

[0082] In some optional embodiments, the association relationship in the above sub-step S23 includes a sequential occurrence relationship and a similar relationship. Among them, if two or more events occur one after another and the occurrence of the latter event depends on the previous event, then the two or more events have a sequential occurrence relationship, that is, they have time correlation. For example, "arriving at xx station" and "dining at xx barbecue restaurant" have a sequential occurrence relationship. If two or more events do not have a sequential occurrence relationship and belong to the same type of events, for example, both belong to dining out events, then the two or more events have a similar relationship. For example, "dining at barbecue restaurant A" and "dining at barbecue restaurant B" have a similar relationship.

[0083] In practice, events in the user memory network can be divided into topics. For example, see Figure 3 , can be divided into travel themes, cultural activities themes, and food themes. Each type of theme forms a subnetwork. The nodes under each subnetwork have both sequential and similar relationships; while the nodes between different subnetworks only have sequential relationships.

[0084] On this basis, the above sub-step S23, based on the association relationship between multiple events, determines the connection mode between nodes, which may further include: for a first event with a sequential occurrence relationship among multiple events, based on the event description information of each first event, determining the occurrence order of each first event. Based on the occurrence order, the first connection mode is used to connect the nodes corresponding to the first event. For a second event with a similar relationship among multiple events, the second connection mode is used to connect the nodes corresponding to the second event. As an example, see Figure 3 The first connection mode may be to use a solid line with an arrow to connect, and the direction of the arrow points to the event that occurs later. The second connection mode may be to use a line without an arrow to connect.

[0085] By using different connection methods to connect the nodes corresponding to events with different association relationships, the type of association relationship between a certain event and another event with which it has an association relationship can be determined intuitively and conveniently, thereby facilitating the subsequent determination of the event sequence and the generation of event sequence description information.

[0086] In some optional embodiments, the above sub-step S24 may further include: for each event in the multiple events, taking the node corresponding to the event in the user memory network as the target node, determining the first node sequence and the second node sequence associated with the target node in the user memory network, and taking the event sequence corresponding to the first node sequence and the event sequence corresponding to the second node sequence as the event sequence associated with the event. The first node sequence is a node sequence starting from the target node and connecting the nodes in a first connection mode, and the second node sequence is a node sequence including the target node and connecting the nodes in a second connection mode.

[0087] As an example, see Figure 3 . For event T1 when the user arrives at Fuzhou Railway Station, after arriving at Fuzhou Railway Station, the user participates in the music festival event A2, then has a dinner at a barbecue restaurant F1, and finally returns to the resident railway station F2, then the first node sequence is {T1->A2->F1->T2}. Similarly, for event T1, its second node sequence is {T1-T2}. For event A2, its second node sequence is {A2-A1-A3}, and for event F1, its second node sequence is {F1-F2-F3-F4}. I will not go into details here.

[0088] The event sequence description information of the above event sequences can be found as follows:

[0089]

[0090] By determining the first node sequence and the second node sequence associated with the target node in the user memory network, the event sequence related to the event occurrence order and the event sequence related to the event type can be obtained, so that rich and comprehensive event sequence description information can be obtained.

[0091] In some optional embodiments, the large language model may be pre-trained. The steps of training the large language model include: first, obtaining the user's historical question information. Then, retrieving the scene description information associated with the historical question information from the scene library. After that, retrieving the user memory information associated with the historical question information from the user memory information library. Finally, the historical question text, the retrieved scene description information, and the retrieved user memory information are used as samples, and the large language model is trained based on the samples.

[0092] It should be noted that for each user, information retrieval is performed from the scene library and user memory information library corresponding to the user, and a large number of samples for the user can be obtained, each sample including {historical question information, scene description information, user memory information}. Each sample is embedded in the input instruction for the large language model, and the input instruction is input into the large language model, which can realize the transformation of the user's personal memory from text form to parameter form, so as to internalize it into the large language model, so that each user can have his own large language model, so that personalized questions and answers can be completed accurately and efficiently based on the large language model.

[0093] It should be noted that in actual applications, data collection, user memory network construction, and reasoning based on a large language model can be performed during the day, and model training and updating can be performed at night. This arrangement ensures that the latest data and information obtained by the model during the day are used for reasoning, and the model is updated and optimized at night to continuously adapt to the personalized needs of users and the changing environment.

[0094] It should be noted that the information generation method provided in the embodiment of the present application can be executed by an information generation device. In the embodiment of the present application, the information generation device provided in the embodiment of the present application is described by taking the information generation method executed by the information generation device as an example.

[0095] like Figure 4As shown, the information generating device 400 of this embodiment includes: a receiving unit 401, used to receive the user's question information; an obtaining unit 402, used to obtain the scene description information of the historical scene associated with the question information and the user memory information associated with the question information, the user memory information includes event description information and event sequence description information, the event description information is used to describe the event associated with the question information, and the event sequence description information is used to describe the event sequence associated with the event; a generating unit 403, used to input the question information, the scene description information and the user memory information into a large language model to obtain the answer information generated by the large language model for the question information.

[0096] In some optional implementations of this embodiment, the acquisition unit 401 is further used to: acquire scene description information of historical scenes associated with the problem information from a scene library; wherein the scene library is generated in advance by the following steps: acquiring basic data of the user in a statistical time period, the basic data including at least one of the following: environmental data, user data, and device data; generating scene description information of multiple historical scenes based on the basic data; and storing the generated scene description information according to set rules to obtain a scene library. By pre-creating a scene library and retrieving scene description information of historical scenes associated with the problem information from the scene library, accurate scene description information of historical scenes associated with the problem information can be easily and quickly acquired.

[0097] In some optional implementations of this embodiment, the generated scene description information is stored according to a set rule to obtain a scene library, including: determining the category of each historical scene in the multiple historical scenes, the category including at least one of the following: life category, work category, and travel category; storing the scene description information of historical scenes belonging to the same category in the multiple historical scenes into the same scene sub-library; and generating a scene library based on the scene sub-libraries corresponding to each category. By storing the scene description information of historical scenes belonging to the same category in the multiple historical scenes into the same scene sub-library, scene matching can be performed conveniently and quickly, thereby efficiently and accurately obtaining the scene description information of the historical scenes associated with the problem information.

[0098] In some optional implementations of this embodiment, the acquisition unit 401 is also used to: acquire user memory information associated with the problem information from a user memory information library; wherein the user memory information library is pre-generated through the following steps: acquiring basic data of the user in a statistical time period, the basic data including at least one of the following: environmental data, user data, and device data; generating event description information of multiple events based on the basic data; using events as nodes of a user memory network, determining connection methods between nodes based on associations between the multiple events, and generating the user memory network based on the nodes and the connection methods; for each of the multiple events, determining an event sequence associated with the event based on the user memory network, and generating event sequence description information of the determined event sequence based on the event description information of each event in the determined event sequence; summarizing the generated event description information of each event and the event sequence description information of each event sequence to obtain a user memory information library. By constructing a user memory information database and obtaining user memory information associated with question information from the user memory information database, the user's memory can be stored in a structured form in text. The stored user memory information can be deeply restored to the user's memory, making the information input into the large language model more extensive and in-depth, so that the answers to the questions generated by the large language model with the help of the above information have higher accuracy and comprehensiveness.

[0099] In some optional implementations of this embodiment, the event description information of multiple events is generated based on the basic data, including: extracting key information of multiple events from the basic data, the key information of each event including at least one of the following: event time, event location, and event name; for each of the multiple events, performing the following steps: obtaining an associated image of the event, the associated image is an image stored at the event time and event location of the event; inputting the associated image into a multimodal large language model to obtain image description information for describing the associated image; based on the key information of the event and the image description information, generating the event description information of the event. By semantically understanding the multivariate data, the event time, event location, event name, and image description information of each event are obtained, so that the summarized event description information is richer and more comprehensive, so that the answers to the questions generated by the large language model with the help of the above event description information have higher accuracy and comprehensiveness.

[0100] In some optional implementations of this embodiment, the association relationship includes a sequential occurrence relationship and a similar relationship; the determination of the connection mode between nodes based on the association relationship between the multiple events includes: for the first event with the sequential occurrence relationship among the multiple events, based on the event description information of each first event, determining the occurrence order of each first event; based on the occurrence order, connecting the nodes corresponding to the first event using a first connection method; for the second event with the similar relationship among the multiple events, connecting the nodes corresponding to the second event using a second connection method. By using different connection methods to connect the nodes corresponding to events with different association relationships, the type of association relationship between a certain event and another event with which it has an association relationship can be determined intuitively and conveniently, thereby facilitating the subsequent determination of the event sequence and the generation of event sequence description information.

[0101] In some optional implementations of this embodiment, for each of the multiple events, based on the user memory network, determining an event sequence associated with the event includes: for each of the multiple events, taking the node corresponding to the event in the user memory network as the target node, determining a first node sequence and a second node sequence associated with the target node in the user memory network, and taking the event sequence corresponding to the first node sequence and the event sequence corresponding to the second node sequence as the event sequence associated with the event; wherein the first node sequence is a node sequence starting from the target node and connecting the nodes in the first connection mode, and the second node sequence is a node sequence including the target node and connecting the nodes in the second connection mode. By determining the first node sequence and the second node sequence associated with the target node in the user memory network, the event sequence related to the order of occurrence of the events and the event sequence related to the event type can be obtained, so that rich and comprehensive event sequence description information can be obtained.

[0102] In some optional implementations of this embodiment, the device further includes a training unit, which is used to: obtain the user's historical question information; retrieve the scene description information associated with the historical question information from the scene library; retrieve the user memory information associated with the historical question information from the user memory information library; use the historical question text, the retrieved scene description information and the retrieved user memory information as samples, and train the large language model based on the samples. In this way, each user can have his or her own large language model, so that personalized questions and answers can be completed accurately and efficiently based on the large language model.

[0103] The device provided by the above-mentioned embodiment of the present application, when receiving the user's question information, first obtains the scene description information associated with the question information and the user memory information associated with the question information, and then inputs the question information, the scene description information and the user memory information into the large language model, thereby obtaining the answer information for the question information generated by the large language model. Among them, the user memory information includes event description information for describing the event associated with the question information and event sequence description information for describing the event sequence associated with the event. Since the information input into the large language model includes, in addition to the question information, the scene description information associated with the question, the event description information of the event associated with the question, and the event sequence description information of the event sequence associated with the event, the above information can deeply restore the user memory related to the question raised by the user, so the information input into the large language model is more extensive and in-depth, so that the answer to the question generated by the large language model with the help of the above information has higher accuracy and comprehensiveness.

[0104] The information generating device in the embodiment of the present application can be an electronic device, or a component in the electronic device, such as an integrated circuit or a chip. The electronic device can be a terminal, or other devices other than a terminal. Exemplarily, the electronic device can be a mobile phone, a tablet computer, a laptop computer, a PDA, a vehicle-mounted electronic device, a mobile Internet device (Mobile Internet Device, MID), an augmented reality (Augmented Reality, AR) / virtual reality (Virtual Reality, VR) device, a robot, a wearable device, a super mobile personal computer (Ultra-Mobile Personal Computer, UMPC), a netbook or a personal digital assistant (Personal Digital Assistant, PDA), etc., and can also be a server, a network attached storage (Network Attached Storage, NAS), a personal computer (Personal Computer, PC), a television (Television, TV), a teller machine or a self-service machine, etc., and the embodiment of the present application is not specifically limited.

[0105] The information generating device in the embodiment of the present application may be a device having an operating system. The operating system may be an Android operating system, an iOS operating system, or other possible operating systems, which are not specifically limited in the embodiment of the present application.

[0106] The information generation device provided in the embodiment of the present application can achieve Figure 1 To avoid repetition, the various processes implemented by the method embodiment are not described here.

[0107] Alternatively, if Figure 5 As shown, an embodiment of the present application also provides an electronic device 500, including a processor 501 and a memory 502, wherein the memory 502 stores programs or instructions that can be executed on the processor 501, and when the program or instructions are executed by the processor 501, the various steps of the above-mentioned information generation method embodiment are implemented, and the same technical effect can be achieved. To avoid repetition, they are not repeated here.

[0108] It should be noted that the electronic devices in the embodiments of the present application include the above-mentioned mobile electronic devices and non-mobile electronic devices.

[0109] Figure 6 A schematic diagram of the hardware structure of an electronic device to implement an embodiment of the present application.

[0110] The electronic device 600 includes but is not limited to: a radio frequency unit 601, a network module 602, an audio output unit 603, an input unit 604, a sensor 605, a display unit 606, a user input unit 607, an interface unit 608, a memory 609, and a processor 610.

[0111] Those skilled in the art will appreciate that the electronic device 600 may also include a power source (such as a battery) for supplying power to each component, and the power source may be logically connected to the processor 610 through a power management system, thereby implementing functions such as managing charging, discharging, and power consumption management through the power management system. Figure 6 The electronic device structure shown in the figure does not constitute a limitation on the electronic device. The electronic device may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently, which will not be described in detail here.

[0112] Among them, the user input unit 607 is used to receive the user's question information; the processor 610 is used to obtain the scene description information of the historical scene associated with the question information and the user memory information associated with the question information, the user memory information includes event description information and event sequence description information, the event description information is used to describe the event associated with the question information, and the event sequence description information is used to describe the event sequence associated with the event; the question information, the scene description information and the user memory information are input into the large language model to obtain the answer information for the question information generated by the large language model.

[0113] Since the information input into the large language model includes, in addition to question information, scene description information associated with the question, event description information of the event associated with the question, and event sequence description information of the event sequence associated with the event, the above information can deeply restore the user's memory related to the question asked by the user. Therefore, the information input into the large language model has greater breadth and depth, so that the answers to the questions generated by the large language model with the help of the above information have higher accuracy and comprehensiveness.

[0114] In some optional implementations of this embodiment, the processor 610 is further configured to obtain scene description information of historical scenes associated with the problem information from a scene library; wherein the scene library is generated in advance by the following steps: obtaining basic data of the user in a statistical time period, the basic data including at least one of the following: environmental data, user data, and device data; generating scene description information of multiple historical scenes based on the basic data; and storing the generated scene description information according to set rules to obtain a scene library. By pre-creating a scene library and retrieving scene description information of historical scenes associated with the problem information from the scene library, accurate scene description information of historical scenes associated with the problem information can be obtained conveniently and quickly.

[0115] In some optional implementations of this embodiment, the processor 610 is further configured to determine the category of each historical scene in the multiple historical scenes, the category including at least one of the following: life category, work category, and travel category; store the scene description information of the historical scenes in the multiple historical scenes belonging to the same category in the same scene sub-library; and generate a scene library based on the scene sub-libraries corresponding to each category. By storing the scene description information of the historical scenes in the same category in the multiple historical scenes in the same scene sub-library, scene matching can be performed conveniently and quickly, thereby efficiently and accurately obtaining the scene description information of the historical scenes associated with the question information.

[0116] In some optional implementations of this embodiment, the processor 610 is also used to obtain user memory information associated with the problem information from a user memory information library; wherein the user memory information library is pre-generated through the following steps: obtaining basic data of the user in a statistical time period, the basic data including at least one of the following: environmental data, user data, and device data; generating event description information of multiple events based on the basic data; using events as nodes of a user memory network, determining connection methods between nodes based on associations between the multiple events, and generating the user memory network based on the nodes and the connection methods; for each of the multiple events, determining an event sequence associated with the event based on the user memory network, and generating event sequence description information of the determined event sequence based on the event description information of each event in the determined event sequence; summarizing the generated event description information of each event and the event sequence description information of each event sequence to obtain a user memory information library. By constructing a user memory information database and obtaining user memory information associated with question information from the user memory information database, the user's memory can be stored in a structured form in text. The stored user memory information can be deeply restored to the user's memory, making the information input into the large language model more extensive and in-depth, so that the answers to the questions generated by the large language model with the help of the above information have higher accuracy and comprehensiveness.

[0117] In some optional implementations of this embodiment, the processor 610 is also used to extract key information of multiple events from the basic data, and the key information of each event includes at least one of the following: event time, event location, and event name; for each of the multiple events, the following steps are performed: obtaining an associated image of the event, the associated image is an image stored at the event time and event location of the event; inputting the associated image into a multimodal large language model to obtain image description information for describing the associated image; based on the key information of the event and the image description information, generating event description information of the event. By semantically understanding the multivariate data, the event time, event location, event name, and image description information of each event are obtained, so that the summarized event description information is richer and more comprehensive, so that the answers to the questions generated by the large language model with the help of the above event description information have higher accuracy and comprehensiveness.

[0118] In some optional implementations of this embodiment, the association relationship includes a sequential occurrence relationship and a similar relationship; the processor 610 is further used to determine the occurrence order of each first event based on the event description information of each first event for the first event with the sequential occurrence relationship among the multiple events; based on the occurrence order, the nodes corresponding to the first event are connected using a first connection method; for the second event with the similar relationship among the multiple events, the nodes corresponding to the second event are connected using a second connection method. By using different connection methods to connect the nodes corresponding to events with different association relationships, the type of association relationship between a certain event and another event with which it has an association relationship can be determined intuitively and conveniently, thereby facilitating the subsequent determination of the event sequence and the generation of event sequence description information.

[0119] In some optional implementations of this embodiment, the processor 610 is further configured to, for each of the multiple events, use the node corresponding to the event in the user memory network as the target node, determine the first node sequence and the second node sequence associated with the target node in the user memory network, and use the event sequence corresponding to the first node sequence and the event sequence corresponding to the second node sequence as the event sequence associated with the event; wherein the first node sequence is a node sequence starting from the target node and connecting the nodes in the first connection method, and the second node sequence is a node sequence including the target node and connecting the nodes in the second connection method. By determining the first node sequence and the second node sequence associated with the target node in the user memory network, the event sequence related to the order of occurrence of the events and the event sequence related to the event type can be obtained, so that rich and comprehensive event sequence description information can be obtained.

[0120] In some optional implementations of this embodiment, the processor 610 is further configured to obtain the user's historical question information; retrieve the scene description information associated with the historical question information from the scene library; retrieve the user memory information associated with the historical question information from the user memory information library; use the historical question text, the retrieved scene description information, and the retrieved user memory information as samples, and train the large language model based on the samples. In this way, each user can have his or her own large language model, so that personalized questions and answers can be completed accurately and efficiently based on the large language model.

[0121] It should be understood that in the embodiment of the present application, the input unit 604 may include a graphics processor (GPU) 6041 and a microphone 6042, and the graphics processor 6041 processes the image data of a static picture or video obtained by an image capture device (such as a camera) in a video capture mode or an image capture mode. The display unit 606 may include a display panel 6061, and the display panel 6061 may be configured in the form of a liquid crystal display, an organic light emitting diode, etc. The user input unit 607 includes a touch panel 6071 and at least one of other input devices 6072. The touch panel 6071 is also called a touch screen. The touch panel 6071 may include two parts: a touch detection device and a touch controller. Other input devices 6072 may include, but are not limited to, a physical keyboard, function keys (such as a volume control button, a switch button, etc.), a trackball, a mouse, and a joystick, which will not be repeated here.

[0122] The memory 609 can be used to store software programs and various data. The memory 609 may mainly include a first storage area for storing programs or instructions and a second storage area for storing data, wherein the first storage area may store an operating system, an application program or instructions required for at least one function (such as a sound playback function, an image playback function, etc.), etc. In addition, the memory 609 may include a volatile memory or a non-volatile memory, or the memory 609 may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), a static random access memory (SRAM), a dynamic random access memory (DRAM), a synchronous dynamic random access memory (SDRAM), a double data rate synchronous dynamic random access memory (DDRSDRAM), an enhanced synchronous dynamic random access memory (ESDRAM), a synchronous link dynamic random access memory (SLDRAM) and a direct memory bus random access memory (DRRAM). The memory 609 in the embodiment of the present application includes but is not limited to these and any other suitable types of memory.

[0123] The processor 610 may include one or more processing units; optionally, the processor 610 integrates an application processor and a modem processor, wherein the application processor mainly processes operations related to an operating system, a user interface, and application programs, and the modem processor mainly processes wireless communication signals, such as a baseband processor. It is understandable that the modem processor may not be integrated into the processor 610.

[0124] An embodiment of the present application also provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the various processes of the above-mentioned information generation method embodiment are implemented and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0125] The processor is the processor in the electronic device described in the above embodiment. The readable storage medium includes a computer readable storage medium, such as a computer read-only memory ROM, a random access memory RAM, a magnetic disk or an optical disk.

[0126] An embodiment of the present application further provides a chip, which includes a processor and a communication interface, wherein the communication interface is coupled to the processor, and the processor is used to run programs or instructions to implement the various processes of the above-mentioned information generation method embodiment, and can achieve the same technical effect. To avoid repetition, it will not be repeated here.

[0127] It should be understood that the chip mentioned in the embodiments of the present application can also be called a system-level chip, a system chip, a chip system or a system-on-chip chip, etc.

[0128] An embodiment of the present application provides a computer program product, which is stored in a storage medium. The program product is executed by at least one processor to implement the various processes of the above-mentioned information generation method embodiment and can achieve the same technical effect. To avoid repetition, it will not be repeated here.

[0129] It should be noted that, in this article, the terms "comprise", "include" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise one..." do not exclude the presence of other identical elements in the process, method, article or device including the element. In addition, it should be pointed out that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in reverse order according to the functions involved, for example, the described method may be performed in an order different from that described, and various steps may also be added, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0130] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a computer software product, which is stored in a storage medium (such as ROM / RAM, a disk, or an optical disk), and includes a number of instructions for a terminal (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present application.

[0131] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the claims, all of which are within the protection of the present application.

Claims

1. A method for generating information, characterized in that: The method comprises: Receive question information input by the user; Acquire scene description information of historical scenes associated with the question information and user memory information associated with the question information, wherein the user memory information includes event description information and event sequence description information, wherein the event description information is used to describe the event associated with the question information, and the event sequence description information is used to describe the event sequence associated with the event; The question information, the scene description information and the user memory information are input into a large language model to obtain answer information for the question information generated by the large language model.

2. The method according to claim 1, characterized in that: Obtaining user memory information associated with the question information, including: Acquire user memory information associated with the question information from a user memory information database; The user memory information database is generated in advance through the following steps: Acquire basic data of the user during a statistical time period, wherein the basic data includes at least one of the following: environmental data, user data, and device data; Based on the basic data, generating event description information of multiple events; Using events as nodes of a user memory network, determining connection modes between nodes based on associations between the multiple events, and generating the user memory network based on the nodes and the connection modes; For each event in the plurality of events, determining an event sequence associated with the event based on the user memory network, and generating event sequence description information of the determined event sequence based on event description information of each event in the determined event sequence; The generated event description information of each event and the event sequence description information of each event sequence are summarized to obtain a user memory information library.

3. The method according to claim 2, characterized in that The generating event description information of a plurality of events based on the basic data includes: Extracting key information of multiple events from the basic data, where the key information of each event includes at least one of the following: time of event, location of event, and name of event; For each of the multiple events, perform the following steps: Acquire an associated image of the event, wherein the associated image is an image stored at the time and location of the event; Inputting the associated image into a multimodal large language model to obtain image description information for describing the associated image; Based on the key information of the event and the image description information, event description information of the event is generated.

4. The method according to claim 3, characterized in that The association relationship includes a sequential occurrence relationship and a similar relationship; the determining of the connection mode between the nodes based on the association relationship between the multiple events includes: For the first events having the sequential occurrence relationship among the multiple events, determining the occurrence order of each first event based on the event description information of each first event; Based on the occurrence order, connecting the nodes corresponding to the first event in a first connection manner; For a second event having the same type relationship among the multiple events, nodes corresponding to the second event are connected using a second connection method.

5. The method according to claim 4, characterized in that For each event of the plurality of events, determining an event sequence associated with the event based on the user memory network comprises: For each event in the multiple events, taking a node in the user memory network corresponding to the event as a target node, determining a first node sequence and a second node sequence in the user memory network associated with the target node, and taking an event sequence corresponding to the first node sequence and an event sequence corresponding to the second node sequence as an event sequence associated with the event; The first node sequence is a node sequence starting from the target node and connecting the nodes in the first connection method, and the second node sequence is a node sequence including the target node and connecting the nodes in the second connection method.

6. An information generating device, characterized in that: The device comprises: A receiving unit, used for receiving question information from a user; an acquisition unit, configured to acquire scene description information of a historical scene associated with the question information and user memory information associated with the question information, wherein the user memory information includes event description information and event sequence description information, wherein the event description information is used to describe an event associated with the question information, and the event sequence description information is used to describe an event sequence associated with the event; A generating unit is used to input the question information, the scene description information and the user memory information into a large language model to obtain answer information for the question information generated by the large language model.

7. The device according to claim 6, characterized in that The acquisition unit is further used for: Acquire user memory information associated with the question information from a user memory information database; The user memory information database is generated in advance through the following steps: Acquire basic data of the user during a statistical time period, wherein the basic data includes at least one of the following: environmental data, user data, and device data; Based on the basic data, generating event description information of multiple events; Using events as nodes of a user memory network, determining connection modes between nodes based on associations between the multiple events, and generating the user memory network based on the nodes and the connection modes; For each event in the plurality of events, determining an event sequence associated with the event based on the user memory network, and generating event sequence description information of the determined event sequence based on event description information of each event in the determined event sequence; The generated event description information of each event and the event sequence description information of each event sequence are summarized to obtain a user memory information library.

8. The device according to claim 7, characterized in that The generating event description information of a plurality of events based on the basic data includes: Extracting key information of multiple events from the basic data, where the key information of each event includes at least one of the following: time of event, location of event, and name of event; For each of the multiple events, perform the following steps: Acquire an associated image of the event, wherein the associated image is an image stored at the time and location of the event; Inputting the associated image into a multimodal large language model to obtain image description information for describing the associated image; Based on the key information of the event and the image description information, event description information of the event is generated.

9. The device according to claim 8, characterized in that The association relationship includes a sequential occurrence relationship and a similar relationship; The determining of the connection mode between the nodes based on the association relationship between the multiple events includes: For the first events having the sequential occurrence relationship among the multiple events, determining the occurrence order of each first event based on the event description information of each first event; Based on the occurrence order, connecting the nodes corresponding to the first event in a first connection manner; For a second event having the same type relationship among the multiple events, nodes corresponding to the second event are connected using a second connection method.

10. The device according to claim 9, characterized in that For each event of the plurality of events, determining an event sequence associated with the event based on the user memory network comprises: For each event in the multiple events, taking a node in the user memory network corresponding to the event as a target node, determining a first node sequence and a second node sequence in the user memory network associated with the target node, and taking an event sequence corresponding to the first node sequence and an event sequence corresponding to the second node sequence as an event sequence associated with the event; The first node sequence is a node sequence starting from the target node and connecting the nodes in the first connection method, and the second node sequence is a node sequence including the target node and connecting the nodes in the second connection method.

Images