Recommendation information generation method and device, storage medium and program product

Abstract

The invention provides a recommendation information generation method and device, a storage medium and a program product. The method comprises the steps that at least one interest point corresponding to a target object is searched based on the target object; determining a first bright spot feature of each point of interest through the first description information of each point of interest; constructing a target prompt word based on the first bright spot feature; and generating recommendation information corresponding to the target object by utilizing the target cue word and a first preset large model. According to the method and the device, the recommendation information which more accurately represents the interested light spots is dynamically provided, diversified requirements of the user are met, and the user experience is enhanced.

Description

Technical Field

[0001] This application relates to the technical field of information processing, and in particular, to a method, apparatus, storage medium, and program product for generating recommendation information. Background Art

[0002] With the popularization of the mobile Internet and intelligent devices, digital map services have become an indispensable part of people's daily lives. Map services not only provide basic navigation and positioning functions, but also provide users with rich geographical information and recommendation services for points of interest (POIs). POIs usually include various locations that users may be interested in, such as restaurants, hotels, shopping malls, gas stations, hospitals, etc.

[0003] POI surrounding recommendation is an important function in map services, which uses the user's current location or POI to recommend nearby services, facilities, or attractions. This function has a wide range of applications in scenarios such as tourism, travel, and business activities. For example, users can view nearby attractions and dining facilities through the POI surrounding recommendation function when traveling, or search for nearby hotels and meeting venues based on the POI surrounding recommendation function during business trips.

[0004] With the continuous improvement of people's living standards and the increasing prosperity of the tourism market, the public's interest in tourism products and travel experiences is growing day by day. The recommendation information provided by the traditional POI surrounding recommendation function is not accurate enough to meet the actual needs of users. Summary of the Invention

[0005] The main purpose of the embodiments of this application is to provide a method, apparatus, storage medium, and program product for generating recommendation information, which realizes dynamically providing more accurate recommendation information that characterizes the highlights of POIs, meets the diverse needs of users, and enhances the user experience.

[0006] In a first aspect, an embodiment of this application provides a method for generating recommendation information, including: searching for at least one POI corresponding to a target object based on the target object; determining a first highlight feature of each POI through first description information of each POI; constructing a target prompt word based on the first highlight feature; and generating recommendation information corresponding to the target object by using the target prompt word and a first preset large model.

[0007] In a second aspect, an embodiment of this application provides a device for generating recommendation information, including:

[0008] A search module, configured to search for at least one POI corresponding to the target object based on the target object;

[0009] A determination module, configured to determine the first highlight feature of each of the points of interest through the first description information of each of the points of interest;

[0010] A construction module, configured to construct a target prompt word based on the first highlight feature;

[0011] A generation module, configured to generate recommendation information corresponding to the target object by using the target prompt word and a first pre-trained large model.

[0012] In a third aspect, an embodiment of the present application provides an electronic device, including:

[0013] At least one processor; and

[0014] A memory communicatively connected to the at least one processor;

[0015] Wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the electronic device to execute the method described in any of the above aspects.

[0016] In a fourth aspect, an embodiment of the present application provides a cloud device, including:

[0017] At least one processor; and

[0018] A memory communicatively connected to the at least one processor;

[0019] Wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the cloud device to execute the method described in any of the above aspects.

[0020] In a fifth aspect, an embodiment of the present application provides a computer-readable storage medium, in which computer-executable instructions are stored, and when a processor executes the computer-executable instructions, the method described in any of the above aspects is implemented.

[0021] In a sixth aspect, an embodiment of the present application provides a computer program product, including a computer program, and when the computer program is executed by a processor, the method described in any of the above aspects is implemented.

[0022] The method, device, storage medium, and program product for generating recommendation information provided by the embodiments of the present application search for the points of interest corresponding to the target object, and determine the first highlight features of the points of interest based on the first description information of each point of interest, so as to extract representative and influential feature information from the first description information. Then, according to the first highlight features of the points of interest, target prompt words are generated, so that the target prompt words can highlight the highlight features of the points of interest and be more directive. When the target prompt words are input into the first preset large model, the first preset large model can accurately generate recommendation information about the points of interest according to the target prompt words. In this way, by utilizing the powerful generation ability of the large model and combining the highlight features of the points of interest, it is possible to dynamically provide recommendation information that can more accurately represent the highlight features of the points of interest, meet the diverse needs of users, and enhance the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The drawings herein are incorporated into the specification and constitute a part of this specification, showing embodiments consistent with the present application, and are used together with the specification to explain the principles of the present application. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0024] Figure 1 It is a schematic structural diagram of an electronic device provided by an embodiment of the present application;

[0025] Figure 2 It is a schematic application scenario diagram of a recommendation information generation system provided by an embodiment of the present application;

[0026] Figure 3 It is a schematic flowchart of a method for generating recommendation information provided by an embodiment of the present application;

[0027] Figure 4 It is a schematic flowchart of a method for generating map recommendation information provided by an embodiment of the present application;

[0028] Figure 5 It is a schematic flowchart of a method for generating recommendation information provided by an embodiment of the present application;

[0029] Figure 6 It is a schematic structural diagram of a recommendation information generation device provided by an embodiment of the present application;

[0030] Figure 7 It is a schematic structural diagram of a cloud device provided by an embodiment of the present application.

[0031] Through the above-mentioned accompanying drawings, specific embodiments of the present application have been shown, and will be described in more detail hereinafter. These drawings and the textual description are not intended to limit the scope of the concept of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. Detailed Description of the Embodiments

[0032] Here, exemplary embodiments will be described in detail, and examples thereof are shown in the drawings. When the following description refers to the drawings, unless otherwise indicated, the same numerals in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application.

[0033] As used herein, the term "and / or" is used to describe the association relationship of associated objects, and specifically represents three possible relationships. For example, A and / or B may represent three cases: A exists alone, A and B exist simultaneously, and B exists alone.

[0034] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data that have been authorized by the user or fully authorized by all parties. Moreover, the collection, use, and processing of relevant data need to comply with relevant laws, regulations, and standards of relevant countries and regions, and corresponding operation entrances are provided for users to choose to authorize or refuse.

[0035] The generation method of the recommended information in the embodiments of the present application can be applied to any technical field involving information recommendation services.

[0036] With the popularization of mobile Internet and intelligent devices, digital map services have become an indispensable part of people's daily lives. Map services not only provide basic navigation and positioning functions, but also provide users with rich geographical information and recommendation services for points of interest (POIs). POIs usually include various locations that users may be interested in, such as restaurants, hotels, shopping malls, gas stations, hospitals, etc.

[0037] POI surrounding recommendation is an important function in map services, which uses the user's current location or POI to recommend nearby services, facilities, or attractions. This function has a wide range of applications in scenarios such as tourism, travel, and business activities. For example, users can use the POI surrounding recommendation function to find nearby attractions, shopping, and dining facilities when traveling, or find nearby hotels and conference venues based on the POI surrounding recommendation function during business trips.

[0038] For example, the user can turn on the POI surrounding recommendation function of the map. When the user is viewing a certain hotel, the map system can display the recommendation information of the points of interest surrounding the hotel on the hotel details page, and the user can view the recommendation information of the points of interest to assist the user in making decisions.

[0039] With the continuous improvement of people's living standards and the increasing prosperity of the tourism market, the public's interest in tourism products and travel experiences is growing day by day. The recommendation information provided by the POI surrounding recommendation function is not accurate enough to meet the actual needs of users. In related technologies, users can use existing large language models to generate summaries and recommendations of surrounding locations.

[0040] However, these solutions have the following problems:

[0041] 1. Unstable generation: The generation result is limited by the content of the surrounding locations and the user's own choices. The materials required for generation need to be manually provided by the user, and the quality of this part of the materials is difficult to be effectively guaranteed.

[0042] 2. Generation quality depends on user prompts: Content generation depends on the user's experience in using large language models, and the generation quality is uneven, unable to ensure consistency.

[0043] 3. The browsing value of the generated content is inconsistent: Content with high browsing value usually requires multiple manual polishings. It is difficult for machine-generated content to maintain consistent browsing value in large-scale maintenance, resulting in large quality differences.

[0044] To solve at least one of the above problems, the embodiments of the present application provide a generation scheme for recommendation information. By searching for the points of interest corresponding to the target object and determining the first highlight features of the points of interest based on the first description information of each point of interest, representative and influential feature information is extracted from the first description information. Then, a target prompt is generated according to the first highlight features of the points of interest, so that the target prompt can highlight the highlight features of the points of interest and be more directive. Inputting the target prompt into the first preset large model can enable the first preset large model to accurately generate recommendation information about the points of interest according to the target prompt. In this way, by utilizing the powerful generation ability of the large model and combining the highlight features of the points of interest, dynamically providing recommendation information that can more accurately represent the highlight features of the points of interest is realized, meeting the diverse needs of users and enhancing the user experience.

[0045] The method of the embodiments of the present application can pre-configure recommendation information for different points of interest in an offline manner, so that different users can see the same recommendation information for the same point of interest.

[0046] Optionally, in an online scenario, real-time generated point-of-interest recommendation information can also be provided to users. For example, user portrait information can be obtained. When a user has a need for point-of-interest information recommendation, it is determined whether the user has a need for real-time generated recommendation information based on the user portrait information. If so, real-time generated point-of-interest recommendation information can be generated online to meet the user's demand for the timeliness of recommendation information.

[0047] The following will describe in detail some embodiments of the present application with reference to the accompanying drawings. Without conflict between the embodiments, the embodiments and the features in the embodiments can be combined with each other. In addition, the step timings in the following method embodiments are only examples and are not strictly limited.

[0048] As Figure 1 shown, this embodiment provides an electronic device 1, including: at least one processor 11 and a memory 12. Figure 1 Taking one processor as an example. The processor 11 and the memory 12 are connected through a bus 10. The memory 12 stores instructions executable by the processor 11. When the instructions are executed by the processor 11, the electronic device 1 can execute all or part of the processes of the methods in the following embodiments to dynamically provide recommendation information that can more accurately represent the highlight features of the point of interest, meet the diverse needs of users, and enhance the user experience.

[0049] In one embodiment, the electronic device 1 can be a mobile phone, a tablet computer, a laptop computer, a desktop computer, or a large computing system composed of multiple computers.

[0050] Figure 2 FIG. 200 is a schematic diagram of an application scenario of a recommendation information generation system provided by an embodiment of the present application. As Figure 2 shown, the system includes: a server 210 and a terminal 220, where:

[0051] The server 210 can be a data platform that provides a recommendation information generation service, such as a map service platform. In an actual scenario, a map service platform may have multiple servers 210. Figure 2 Taking 1 server 210 as an example.

[0052] The terminal 220 can be a mobile device used by a user to log in to the map service platform, such as a computer, a mobile phone, a tablet, etc. that establish a communication connection with the map service platform. There can also be multiple terminals 220. Figure 2 Taking 2 terminals 220 as an example for illustration.

[0053] Information can be transmitted between the terminal 220 and the server 210 via the Internet so that the terminal 220 can access the data on the server 210. The above-mentioned terminal 220 and / or server 210 can both be implemented by the electronic device 1.

[0054] The recommendation information generation scheme of the embodiments of the present application can be deployed on the server 210, or on the terminal 220, or partially on the server 210 and partially on the terminal 220. In an actual scenario, it can be selected based on actual needs, and this embodiment does not make any limitations.

[0055] When the recommendation information generation scheme is fully or partially deployed on the server 210, a call interface can be opened to the terminal 220 to provide algorithm support to the terminal 220.

[0056] The method provided by the embodiments of the present application can be implemented by the electronic device 1 executing corresponding software code and by interacting with the server for data. Among them, the electronic device 1 can be a local terminal device. When the method runs on the server, the method can be implemented and executed based on a cloud interaction system, where the cloud interaction system includes a server and a client device.

[0057] In a possible implementation manner, the method provided by the embodiments of the present application provides a graphical user interface through a terminal device, where the terminal device can be the aforementioned local terminal device or the client device in the aforementioned cloud interaction system.

[0058] Please refer to Figure 3 , which is a method for generating recommendation information according to an embodiment of the present application. This method can be executed by the Figure 1 shown electronic device 1 and can be applied to the Figure 2 shown recommendation information generation application scenario, so as to not only dynamically provide more attractive and practical point-of-interest information recommendations to meet the diverse needs of users, but also improve the relevance and accuracy of the recommendation information to the target object and enhance the user experience. Taking the terminal 220 as an execution end as an example, this method includes the following steps:

[0059] Step 301: Search for at least one point of interest corresponding to the target object based on the target object;

[0060] In this step, the target object is the object currently queried by the user, such as a shopping mall, hotel, hospital, scenic spot, school, etc. that the user is viewing. The target object can also be a location, such as the location where the user is currently located, or the target location selected by the user on the map. A point of interest is an object existing at a location related to the target object, such as, but not limited to, a shopping mall, hotel, hospital, scenic spot, school, etc. within a certain range of the target object. When the target object of the user's concern is determined, one or more relevant points of interest can be automatically searched for the target object based on map data.

[0061] In one embodiment, step 301 may specifically include: determining at least one associated candidate category based on the category of the target object; taking the target object as the center point and searching for at least one point of interest under the candidate category within a preset range.

[0062] In this embodiment, when searching for points of interest for the target object, first identify the target category to which the target object belongs, and further determine at least one candidate category associated with the target category. Then, taking the target object as the center point, search for at least one point of interest under these candidate categories within a preset range. In this way, not only the accuracy of the search is improved, but also the breadth of the search is expanded, enabling the user to discover potential points of interest within a more relevant category range. Through this associative search strategy, the system can more effectively capture the potential interests and needs of the user, provide richer and more personalized recommendation results, and thus significantly improve the user experience and satisfaction.

[0063] For example, if the user triggers a surrounding recommendation instruction while currently located in residential community A, the target object is the location where community A is located. First, identify that the target category of community A is a residential community, and then search for candidate categories associated with residential communities in the preset database. The preset database stores the association relationships of multiple categories. Candidate categories are the categories that the user may be interested in when in the location of the target category. For example, for the category of residential community, its associated candidate categories include, but are not limited to, hospitals, education, fitness, catering, etc. There may be one or more points of interest under each candidate category. For example, the education category includes 2 schools, and the catering category includes 10 restaurants. The preset range can be set according to actual needs. For example, it can be the default preset radius range of the system (such as within a radius of 2 kilometers), or the user can customize the preset range to meet the user's personalized needs. It can also be determined according to big data analysis of the user's query habits. For example, based on map data, taking community A as the center point, 1 hospital, 2 schools, and 10 restaurants are searched within a range of 2 kilometers, for a total of 13 points of interest.

[0064] Optionally, for the multiple points of interest searched in step 301, preliminary filtering of redundant data can be performed, such as filtering out points of interest with high homogeneity and / or filtering out points of interest with negative reviews, to improve the quality of the points of interest recommended subsequently.

[0065] Step 302: Determine the first highlight feature of each point of interest through the first description information of each point of interest.

[0066] In this step, the first description information is used to characterize the basic information of the point of interest, such as the text description information of the point of interest, which generally includes information such as the basic introduction of the point of interest and user reviews. The highlight feature is used to characterize the experience characteristics of the point of interest in different dimensions for the user. For example, the highlight features of a food and beverage point of interest include, but are not limited to, features in dimensions such as the in-store dining environment, the taste of the food, and the dining service, which can prominently represent the experience characteristics of the food and beverage point of interest for the user. The content of the first description information is often relatively large, and it can be summarized and refined to determine the first highlight feature of the point of interest. For example, by summarizing the high-quality reviews of each point of interest, the summary information of the highlights obtained can be used as the corresponding first highlight feature. The first highlight feature can not only accurately provide information on the impact of the point of interest on the user experience, facilitating the provision of an accurate data basis for the generation of subsequent target prompt words, but also reduce the data volume and improve the data processing efficiency.

[0067] In one embodiment, step 302 may specifically include: obtaining the first description information of each point of interest, where the first description information includes the evaluation information and / or introduction information of the point of interest; inputting the first description information into a second preset large model so that the second preset large model generates the first highlight feature of each point of interest according to preset prompt words.

[0068] In this embodiment, by obtaining the first description information of each point of interest, the first description information is used to characterize the basic information of the point of interest, including but not limited to the evaluation information and / or introduction information of the point of interest. For example, for a food and beverage point of interest, its description information may include text information such as customer evaluations of the food, merchant introductions of the food, and basic information of the food itself. And these description information are input into the second preset large model, so that the second preset large model generates the first highlight feature of the corresponding point of interest according to the preset prompt words. By utilizing the natural language processing ability of the large model, the highlight information that best represents the characteristics of the point of interest can be extracted from the rich first description information. Here, the highlight information refers to the characteristics and advantages information of the POI, including but not limited to POI features, POI core information, POI user experience, etc., thereby providing accurate target prompt words for the generation of subsequent recommendation information.

[0069] Optionally, the above preset prompt words can be set according to actual needs. For example, the preset prompt words can be as follows:

[0070] "If you are a tour guide. Please carefully read the following reviews and complete the writing of the highlight recommendations for + place_type according to the following template, which requires summarization and refinement."

[0071] + "Highlight recommendations for + place_type, requiring summarization and refinement."

[0072] "Please strictly ensure the authenticity and accuracy of the information, and indicate the review entries referred to when recording each piece of information."

[0073] + ("You can extract at most one highlight." if review_numbers <= 2 else

[0074] "You can extract at most two highlights." if 2 < review_numbers <= 5 else

[0075] "You can extract at most four highlights.") +

[0076] "Please try to extract the highlight content that + user_type most expects to get."

[0077] "Please summarize and refine according to the content mentioned in the reviews, and omit the corresponding parts for aspects not mentioned."

[0078] "The review content is as follows:\n" + top10_reviews + "\n"

[0079] "Please complete the summary of the highlight recommendations for + poi_name according to the following framework:\n"

[0080] "As a tour guide, extracting valuable information from the reviews of + user_type is an important task."

[0081] "Because these reviews can help me better recommend the highlight information of + place_type,"

[0082] "and provide useful suggestions for future + user_type.\n"

[0083] "-- Pay more attention to answering only based on the data in the provided reviews, and avoid introducing unsubstantiated details. Be sure not to give speculative information based on the reviews.\n"

[0084] "-- Answer strictly based on the provided review data\n"

[0085] "-- Remove the AI flavor from the answer\n"

[0086] "--Integrity: The answer should be a direct reflection of the original comment information, rather than an extension or inference based on additional sources.\n"

[0087] "--Consistency: The information between the answer and the original comment should be consistent, avoiding the introduction of viewpoints or details not mentioned.\n"

[0088] "## Highlights Recommendation of " + poi_name + "\n"

[0089] Among them, place_type represents the category of the point of interest, user_type represents the user type, and poi_name represents the name of the point of interest. This preset prompt combines the description information of the POI itself, including: the user type the POI faces, the POI type, the number of POI comments, etc. By presetting the prompt to design the input structure of the second preset large model, the input vector can better interact with the model vector space, so that the output of the second preset large model can be more in line with expectations.

[0090] Optionally, in addition to using the large model to generate the first highlight feature of the point of interest, one or more of the following methods can also be used to generate the first highlight feature:

[0091] A. Natural language processing technology can be used to generate the first highlight feature of the point of interest. For example, text summarization technology can be adopted. Text summarization technology is a natural language processing task that uses the model architecture of ordinary neural networks and aims to extract key information from a large amount of text to generate a concise summary. The first description information corresponding to the point of interest can be input into the text summarization model, and the key information output by the text summarization model according to the first description information can be used as the highlight feature of the corresponding point of interest.

[0092] B. For some key points of interest, the first highlight feature of the relevant point of interest can also be obtained by manual or crowdsourcing input.

[0093] In the actual scenario, considering that large models generally have problems such as attention loss, poor task performance, and loss of task format in long text modeling, in this embodiment, the description information of surrounding POIs is compressed in one stage, and a large amount of comment information content and / or introduction information is compressed into the highlight feature information of the POI itself, so as to improve the performance of the first preset large model in the recommendation information generation task and improve the stability of the recommendation information.

[0094] Step 303: Construct a target prompt based on the first highlight feature.

[0095] In this step, prompt construction refers to constructing relevant prompts applied to the large language model based on the input keywords or sentences to help the large language model better understand and express its own meaning. Aiming at the problem that the quality of recommended information in related technologies depends on user prompts, in this embodiment, by using the first highlight feature of the point of interest, more targeted target prompts are automatically constructed without relying on user-input prompts, which not only simplifies user operations and improves the user experience, but also improves the accuracy of the target prompts.

[0096] In one embodiment, the method further includes: determining the second highlight feature of the target object based on the second description information of the target object; step 303 may specifically include: constructing a target prompt based on the first highlight feature of the point of interest and the second highlight feature of the target object.

[0097] In this embodiment, the second description information is used to characterize the basic information of the target object, including but not limited to evaluation information and / or introduction information of the target object. For example, for a food and beverage target object, its second description information may include text information such as customer evaluations of the food, merchant introductions of the food, and basic information of the food itself. The method for determining the second highlight feature of the target object can refer to the determination process of the first highlight feature of the point of interest above. These second description information can be input into the second preset large model, so that the second preset large model generates the second highlight feature of the target object according to the preset prompt. By using the natural language processing ability of the large model, the highlight information that best represents the characteristics of the target object can be extracted from the rich second description information. Here, the highlight information includes but is not limited to the characteristic information of the target object, the core information of the target object, the user experience of the target object, etc., so as to provide accurate target prompts for the generation of subsequent recommended information.

[0098] Optionally, if the target object itself has rich enough description information (such as comments, ratings, etc.) in the actual scenario, the second highlight feature of the target object can be determined based on this part of the description information. For example, well-known hotels, restaurants and other objects generally have very rich description information. If the target object itself does not have rich enough description information, relevant surrounding information can be combined as the second description information to determine the second highlight feature of the target object. For example, a hotel near a famous scenic spot will tend to emphasize the advantages of this hotel for visiting the scenic spot.

[0099] In the process of constructing the target prompt, the highlight features of the target object itself and the highlight features of the corresponding point of interest can be comprehensively considered, so that the target prompt not only has directivity to the point of interest, but also has directivity to the target object, and can provide more target prompts that meet the actual needs, so as to make the recommended information more complete and substantial.

[0100] In one embodiment, constructing a target prompt based on the first highlight feature of the point of interest and the second highlight feature of the target object includes: calculating the recommendation score of each point of interest according to a preset scoring dimension; for points of interest of the same category, sorting multiple points of interest from largest to smallest according to the recommendation score; determining at least one point of interest ranked in the top preset number as the point of interest to be recommended under the corresponding category; generating a target prompt based on the first highlight feature of the point of interest to be recommended, the second highlight feature of the target object, and a preset prompt structure; wherein the preset prompt structure includes a feature slot about the point of interest to be recommended and / or a feature slot about the target object.

[0101] In this embodiment, when there are multiple points of interest, if the points of interest are not sorted and there is no sorting result of the points of interest in the constructed target prompt, the first preset large model needs to sort the points of interest to be recommended by itself subsequently. In the actual scenario, there are no obvious rules and constraints for the large model to sort restaurants. In most cases, it does not meet the intention of the user side. For example, the user side may prefer restaurants that are closer and have higher ratings, while the sorting result of the large model by itself often judges completely according to the highlight content of the points of interest to be recommended. Therefore, if there are multiple points of interest corresponding to the target object, these points of interest can be screened to improve the quality of the points of interest. First, calculate the recommendation score of each point of interest according to the preset scoring dimension, and for points of interest of the same category, sort them from largest to smallest according to these recommendation scores. Then determine at least one point of interest ranked in the top preset number as the point of interest to be recommended under the corresponding category. The preset number can be set according to actual needs. For example, taking Community A as the center point, 1 hospital, 2 schools, and 10 restaurants are searched within a range of 2 kilometers, a total of 13 points of interest. If according to the scoring result of the recommendation score, the scores of 2 restaurants are lower than the set value, then these 2 restaurants with low scores are removed, and the remaining 11 points of interest are sorted under their respective categories. For the medical category, the hospital ranked first can be selected as the point of interest to be recommended under the medical category. For the catering category, the top 3 restaurants can be selected as the points of interest to be recommended under the catering category, which not only ensures that the points of interest to be recommended have high relevance and attraction among the same category, but also avoids recommending points of interest with poor ratings to users. Then, according to the second highlight feature of Community A, the first highlight feature of the points of interest to be recommended under each category, and the preset prompt structure, generate a target prompt. In the case of having a sorting result, the sorting result and order will be directly marked in the target prompt. Taking the sorting result of restaurants as an example, the first preset large model will directly generate and sort the recommended information according to the sorting of the restaurants in the target prompt. Using the target prompt with a sorting result to guide the large model makes the generated recommended information more inclined to the needs of the user side.

[0102] In one embodiment, the preset scoring dimensions include one or more of the highlight feature dimension, the distance dimension, and the information quantity dimension. Calculating the recommendation score for each point of interest according to the preset scoring dimensions includes: for each point of interest, calculating the feature score of the first highlight feature of the point of interest according to the scoring rule of the highlight feature dimension. Determining the distance score of the point of interest according to the distance between the point of interest and the target object. Determining the information quantity score of the point of interest according to the information quantity of the first highlight feature. Determining the recommendation score of the point of interest according to the feature score, the distance score, the information quantity score, and the preset weight.

[0103] In this embodiment, the preset scoring dimensions are used to score the information quality of the points of interest. Different scoring dimensions can represent the influencing factors of the points of interest on the user experience from different perspectives, so as to screen out the points of interest with better information quality based on the scoring results and recommend them to the user. The preset scoring dimensions can be set according to actual needs, such as including but not limited to the highlight feature dimension, the distance dimension, and the information quantity dimension, which are used to represent the experience that the points of interest can bring to the user in terms of highlight features, distance, information quantity, etc. Different scoring rules can be configured for each scoring dimension. For example, the scoring rule of the highlight feature dimension can be: determining the feature score according to the scoring data given by the user in the historical evaluation information. The scoring rule of the distance dimension can be that the score decreases non-linearly as the distance between the point of interest and the target object increases. The scoring rule of the information quantity dimension can be that the score increases linearly as the information quantity of the highlight feature increases.

[0104] By comprehensively considering multiple preset scoring dimensions, the recommendation score of each point of interest is calculated. For each point of interest, first, according to the scoring rule of the highlight feature dimension, calculate the feature score of its first highlight feature, then, according to the distance between the point of interest and the target object and the scoring rule of the distance dimension, determine its distance score, and optionally, according to the information quantity of the first highlight feature and the scoring rule of the information quantity dimension, determine the information quantity score of the point of interest. Finally, by combining the feature score, the distance score, and the information quantity score, and combining the preset weight, determine the total recommendation score of the point of interest. This scoring mechanism not only comprehensively evaluates the multi-faceted characteristics of the points of interest, but also ensures the objectivity and rationality of the recommendation results, enabling users to obtain more accurate and personalized recommendations of points of interest and improving the overall user experience.

[0105] For example, a. The calculation method of the recommendation score corresponding to the point of interest for the target object can be as follows:

[0106] i. Calculate the highlight score F1 of the point of interest itself (i.e., the feature score).

[0107] ii. Calculate the distance score F2 of the point of interest from the target object. The rule is: it decreases non-linearly with the distance. Optionally, the distance score formula can be as follows:

[0108]

[0109] Among them, D is the distance between the point of interest and the target object, in kilometers. The distance score of the point of interest decreases non-linearly with the distance D. When D is 0, the distance score F2 is 8 points. When the distance D is 4 kilometers, the distance score F2 is 0 points. When D is even farther, it becomes a negative score. For different actual scenarios, it is allowed to adjust the corresponding weights. The non-linear design mainly considers the needs of users in the actual scenario: when the distance is close enough, especially within the walking distance, users will be more sensitive to the distance. 100 meters is significantly better than 200 meters. However, when the distance exceeds a certain value, the sensitivity of users decreases, and the difference between 1 kilometer and 1.1 kilometers is not obvious.

[0110] iii. Calculate the information quantity score F3 of the point of interest. The rule is: it increases linearly with the information quantity of the highlight features, such as increasing linearly with the number of highlights hit by the point of interest itself.

[0111] iv. Calculate the recommendation score corresponding to the point of interest by combining the preset weights = 0.5 * F1 + 0.25 * F2 + 0.25 * F3.

[0112] In one embodiment, the preset prompt word structure includes a feature slot about the point of interest to be recommended and / or a feature slot about the target object.

[0113] In this embodiment, the preset prompt word structure includes but is not limited to a feature slot about the point of interest to be recommended and a feature slot about the target object. The feature slots in the prompt word structure are used to fill in the corresponding feature description information. For example, the feature slot of the point of interest to be recommended is used to fill in the relevant feature description information based on the first highlight feature of the point of interest to be recommended during the process of generating the target prompt word. Here, the feature description information includes but is not limited to the self-information of the point of interest to be recommended, the distance from the target object, tags, highlight information, etc. The feature slot of the target object is used to fill in the relevant feature description information based on the second two-point feature of the target object during the process of generating the target prompt word. Here, the relevant feature description information of the target object includes but is not limited to the self-information of the target object, its own tags, its own highlight information, etc. Through the structured prompt word generation mechanism, the accuracy and personalization degree of the target prompt word are improved, and then the target prompt word can more effectively guide the large model to focus on the most valuable points of interest, thereby optimizing the recommendation information.

[0114] For example, the configuration of the preset prompt word structure can be combined with the following content:

[0115] i. The self-information, self-labels, and self-highlight information of the target object. Taking Hotel B as an example of the target object, the self-information of Hotel B includes, but is not limited to, the historical evaluation information of users on Hotel B and the product information related to Hotel B (such as room introduction, breakfast categories, etc.). Based on the historical reviews, products, and other information of Hotel B itself, the highlights of Hotel B can be understood, such as its superior geographical location, bustling surroundings, high cost performance, etc. Labels mainly refer to the industry labels of the target object. For example, Hotel B is an economy hotel, and some hotels contain label information such as free parking.

[0116] ii. The self-information of the interest point to be recommended, the distance from the target object, labels, and highlight information.

[0117] iii. It is required to be expressed in a human - touch way with rich rhetoric.

[0118] iv. Remove duplicate brands and eliminate large - model hallucinations.

[0119] v. Requirements for generating prompt words: Ensure the correct format and labels for subsequent matching.

[0120] Taking Xinglong Wenquan Jinri Hotel as an example of the target object, the constructed target prompt words can be as follows:

[0121] Please assume that you are a tour guide aiming to promote Xinglong Wenquan Jinri Hotel and its surrounding environment. Next, I will provide you with relevant materials about Xinglong Wenquan Jinri Hotel. Please first briefly introduce Xinglong Wenquan Jinri Hotel, and then introduce different brands and locations in each category in the given order. The following are the relevant materials:

[0122] The introduction of scenic spots is as follows:

[0123] Name: Bali Village

[0124] Distance: 3.6 kilometers

[0125] Sorting: 1

[0126] Labels: Scenic spot; Scenic area; 3A scenic area

[0127] Summary of highlights: ## Bali Village highlight recommendations

[0128] Highlight 1: A photo - taking and punching - card place with exotic customs

[0129] **Details**: Bali Village is famous for its unique Balinese style in Indonesia, offering multiple popular photo-taking spots, including the Sky Gate at the scenic area entrance and the elaborately arranged Southeast Asian-style buildings and plant backgrounds. Visitors can rent ethnic or Indonesian-style clothing (reference review: starting from 30 yuan), paired with headgear and earrings, and experience full makeup services (80 yuan) to take beautiful photos with exotic styles within the scenic area. Without having to travel far to Southeast Asia, visitors can enjoy the fun of taking photos, especially being able to take satisfactory photos at the scenic area entrance (reference review entries: 1, 4, 5, 6, 7, 8).

[0130] Highlight 2: In-depth experience of the culture of returned overseas Chinese from Southeast Asia

[0131] **Details**: Bali Village is not only a great place for taking photos but also a window to showcase the culture of returned overseas Chinese from Southeast Asia. Visitors can learn about the living situations of early returned overseas Chinese after they came back to China and the development and changes of the overseas Chinese farm over the past 60-odd years. By visiting the Southeast Asian Cultural Hall, they can feel the precipitation of history and the integration of cultures (reference review entries: 6, 7, 8).

[0132] Highlight 3: Relaxing tour of the tropical botanical garden

[0133] **Details**: The tropical botanical garden around Bali Village provides visitors with rich natural landscapes and leisure options. In the botanical garden, visitors can admire a wide variety of tropical plants, such as pepper trees, jackfruit, etc., learn about their special cultivation methods, and at the same time enjoy the quiet, cool environment and fresh air filled with the strong fragrance of flowers, which is very suitable for a leisurely stroll (reference review entries: 6, 7).

[0134] Highlight 4: Special food and shopping experiences

[0135] **Details**: In addition to taking photos and cultural experiences, Bali Village and its surrounding areas also offer opportunities to taste authentic Southeast Asian cuisine. Visitors can enjoy local Indonesian-style dishes, especially the renowned coffee and special snacks. Moreover, there are several specialty stores within the scenic area and its vicinity, providing opportunities to buy special souvenirs to enrich the travel experience (reference review entries: 5, 6, 8).

[0136] In summary, Bali Village not only attracts visitors with its exotic style but also offers diverse experiences of deeply understanding the culture of returned overseas Chinese from Southeast Asia, leisurely experiencing the tropical botanical garden, and tasting special foods, making it one of the highlights not to be missed in Wanning's tourism.

[0137] Name: Xinglong Tropical Botanical Garden

[0138] Distance: 3.9 kilometers

[0139] Sorting: 2

[0140] Tags: Scenic Spot; Scenic Area; 4A Scenic Area|Scenic Spot; Botanical Garden|Scenic Spot; Natural Landscape

[0141] Highlights Summary: ##Highlights Recommended for Xinglong Tropical Botanical Garden

[0142] Highlight 1: Encyclopedia of Tropical Plants and a Photography Paradise

[0143] **Content Extraction**: Xinglong Tropical Botanical Garden has become a paradise for plant lovers with its rich variety of tropical plants (more than 2,300 species) and is known as an encyclopedia of tropical plants. The scenery in the garden is charming, with coconut trees contrasting beautifully against the clear sky, making it an excellent choice for photography enthusiasts. Reference Comment:

The tropical rainforest is so photogenic!! It's extremely photogenic with a clear sky and coconut trees

[0144] **Information Source**: The mention of "extremely photogenic" and "various posing spots and locations" indicates that the scenery is pleasant and suitable for taking photos.

[0145] Highlight 2: Rich Experiences and Free Tastings

[0146] **Content Extraction**: In addition to viewing plants, visitors can also participate in experience projects such as making chocolate and perfume, which is suitable for family outings with children. The park provides free guided tours, making it suitable for family outings, especially those with children and the elderly. Moreover, visitors can enjoy free tastings of Xinglong coffee, cocoa series beverages, and famous teas produced in the park at the rest area. Reference Comment:

Free tastings are mentioned in the ticket section

[0147] **Information Source**: It includes "experience projects of making chocolate and perfume by oneself", "free tastings at the park's rest reception and tasting area", and "free docents in the park".

[0148] Highlight 3: Nature Education and Affordable Prices

[0149] **Content Extraction**: The botanical garden is an excellent place for nature education. It not only has a wide variety of plants but also provides detailed guided tour services, making it suitable for visitors of all ages to learn about tropical plant knowledge. The ticket price is affordable, and there are discounts for specific groups such as teachers. Reference Comment:

Ticket information and the fact that using a teacher's certificate can get a 50% discount

[0150] **Information Source**: The mention of "this small but amazing botanical garden in Wanning" and "using a teacher's certificate can get a 50% discount!!" shows that the price is reasonable and there are preferential policies.

[0151] Highlight 4: In - depth Natural Experience and Professional Explanation

[0152] **Content Refinement**: Xinglong Tropical Botanical Garden offers opportunities for in-depth contact with nature. Visitors can immerse themselves in the "Tropical Kingdom" and experience the rich species of the tropical rainforest. Professional docent services, such as Xiaoming, Xiaohua, Xiaogang, etc., enable visitors to gain a deep understanding of tropical plants and enhance the tour experience. Reference Comments: [Positive reviews on docent services mentioned]

[0153] **Information Source**: It includes "Sister Xiaohua who explained to us was very serious and responsible" and "Guide Xiaoming's explanation was very detailed", indicating that the docent service is professional and well-received.

[0154] Precautions

[0155] **Information Source**: [Avoiding Pit Guide for Wanning Xinglong Botanical Garden: Don't Touch Randomly!]

[0156] **Content Refinement**: Be sure to bring mosquito repellent as there are many mosquitoes in the park. Also, due to the large variety of plant species, avoid touching unknown plants randomly.

[0157]

Omitted highlights

[0158] Please first give a brief introduction to Jinri Hotel in Xinglong Hot Spring, and then select different brands of locations for each category, with each location expressed in one sentence highlighting its most prominent highlight or feature. For the food category, please prioritize introducing special dishes or signature dishes.

[0159] The following is the template for the promotional page. Please output it in the following JSON (JavaScript Object Notation, a file format and data exchange format with an open standard) format: (The specified output format is configured here).

[0160] The solution of the embodiment of this application, based on the sorted surrounding points of interest and their highlight information, can generate targeted target prompt words. Then, when generating recommended text, it can comprehensively consider factors such as the target object's own information, its own tags, its own highlight information, the information of the point of interest itself, distance, tags, highlight information, etc., and output recommended information about the surrounding points of interest that is full of humanity and rich in rhetoric.

[0161] Step 304: Use the target prompt words and the first preset large model to generate recommended information corresponding to the target object.

[0162] In this step, the target prompt is generated based on the first highlight feature of the interest points to be recommended, which accurately describes the relevant highlight information of the interest points to be recommended. Therefore, the target prompt is input into the preset large model so that the preset large model can accurately identify and highlight the unique selling points and advantages of the interest points to be recommended according to the target prompt, and provide more attractive and informative recommendation information for users. This not only improves the efficiency and accuracy of generating recommendation information, enhances the user experience and satisfaction in the selection and decision-making process, but also reduces the user's usage threshold and improves the user interaction experience because it does not rely on the user to manually input prompts.

[0163] Here, the recommendation information can be recommendation text, images, audio and video, etc. for the interest points to be recommended, and multiple types of recommendation information can be combined to further improve the user experience.

[0164] The first preset large model for generating recommendation information and the second preset large model for generating the target prompt can be the same or different. For example, according to different functional requirements, different samples are used to fine-tune the large model to obtain a large model that can perform well on different tasks. The embodiments of the present application do not make any limitations in this regard. In one embodiment, after step 304, it further includes: detecting the fields of the recommendation information and correcting the error field information in the recommendation information.

[0165] In this embodiment, the content with high browsing value in the actual scenario usually needs to go through multiple manual polishings, and the content automatically generated by the large model may be difficult to maintain consistent browsing value in large-scale maintenance, resulting in large quality differences. To address this issue, after generating the recommendation information about the interest points, this embodiment can also perform post-processing on the generated recommendation information to ensure the consistency of content browsing. For example, the step of detecting the fields of the recommendation information generated in step 304 can be performed, including but not limited to verifying the fields of the generated text, detecting missing information, etc. The process can be as follows:

[0166] 1. For example, if the recommendation information uses the JSON data format, a JSON format verification tool can be used to verify the JSON format of the recommendation information to ensure the data correctness and consistency of the recommendation information.

[0167] 2. Supplementary of factual information: It refers to supplementing or correcting relevant facts according to the information provided by the user to ensure the accuracy and integrity of the information. For the interest points to be recommended that lack factual information, the factual information can be supplemented and corrected. The factual information includes but is not limited to the distance information of the interest points to be recommended, the name of the interest points to be recommended, etc.

[0168] 3. Filter bug (defect) fields: This refers to screening out fields containing bug information from the data to help locate and resolve problems more quickly. In this embodiment, bug fields in the recommended information can be filtered out. For example, the bug fields can be text fields with incorrect formats generated by a preset large model, and they can be corrected.

[0169] Through this process, the system can automatically identify possible incorrect field information in the recommended information and make necessary corrections. This post-processing mechanism not only maintains the consistent browsing value of the recommended information during large-scale maintenance but also improves the accuracy and reliability of the recommended information, ensuring that the information received by users is clear and error-free. By detecting and correcting the fields of the recommended information, it effectively reduces user confusion or misguidance that may be caused by information errors, further optimizing the user experience and the quality of information services.

[0170] In one embodiment, after step 304, it further includes: evaluating the output result of the first preset large model according to the recommended information, and adjusting the first preset large model according to the evaluation result.

[0171] In this embodiment, after generating the recommended information about the point of interest, the output result of the first preset large model can be further evaluated, and the first preset large model can be adjusted according to the evaluation result. By systematically evaluating the quality, relevance, and user satisfaction of the recommended information, the deficiencies in the model output are identified and corresponding optimization adjustments are made. Through this feedback mechanism, the system can continuously improve the performance and output effect of the large model, making the generated recommended information more in line with user needs and expectations. This dynamic adjustment process not only improves the accuracy and practicality of the recommended information but also enhances the adaptive ability of the system, ensuring that users always obtain a high-quality service experience in a changing environment.

[0172] In this embodiment, large model scoring refers to using a large model to evaluate and score a certain task or problem to help users better understand and evaluate the results. The first preset large model can be scored according to the output recommended information. For example, the evaluation process includes but is not limited to the following:

[0173] a. Score the recommended information about each point of interest to be recommended output by the first preset large model to determine the selection score of the point of interest to be recommended. The selection score of the point of interest to be recommended can be used to judge whether the large model can accurately summarize the relevant content of the point of interest to be recommended and select appropriate highlight information.

[0174] b. Information correctness score: Evaluate the correctness and accuracy of the recommended information output by the first preset large model, usually including aspects such as the accuracy of facts and the rationality of logic. It is used to judge whether the large model has selected the correct and important information in the point of interest to be recommended.

[0175] c. Overall highlight score: It refers to the overall highlights and advantages of evaluating recommended information, usually including aspects such as user experience and information richness. It can be evaluated by reading the text of the generated recommended information to see if the text itself has sufficient highlights.

[0176] The above method for generating recommended information first generates highlight information using point-of-interest related information, and then proceeds with the link for generating recommended information. This link can automatically produce recommended information and perform rule-based post-processing to supplement factual information for the verified fields of the generated text. It can utilize the powerful generation ability of the large language model, combined with the surrounding information of the location selection, to accurately, safely, and comprehensively introduce the surrounding information of the target object with fluent language expression, which can improve the information value of the target object itself. This method can be applied in batches to a large number of POIs, greatly shortening the cycle of producing surrounding recommendation texts, improving the browsing information benefits of POIs, and enhancing the user information experience. It can also use the large model for batch scoring and evaluation. Without manual intervention, this link can achieve batch evaluation of relevant content and can further detect and limit mass-produced texts safely and accurately.

[0177] The embodiments of this application at least have the following beneficial effects:

[0178] 1. Aiming at the problem of unstable generated content of the large model in related technologies, this problem is solved by specifically designing a two-stage framework. In the first stage, a round of comment summary is performed on the points of interest around the target object to extract the most highlighted and expressive parts of the surrounding points of interest. Considering that large models generally have problems such as attention loss, poor task performance, and loss of task format in long text modeling, the information of the surrounding points of interest is compressed in the first stage, and a large amount of comment content is compressed into the highlights of the points of interest themselves to improve the performance ability of the large model in this task.

[0179] 2. Aiming at the problem that the quality of the generated content of the large model in related technologies depends on user prompt words, this problem is solved by designing targeted text prompt words. By presetting the prompt word structure and framework, a round of comment summary is performed on the points of interest around the target object in the first stage to extract the most highlighted and expressive parts of the surrounding points of interest. In the second stage, the surrounding recommendation text is further obtained using the highlight text (i.e., highlight features) generated in the first stage. By enriching the prompt words, the generation accuracy and diversity of the model itself are improved, and for some cases with problems during generation, they are filtered by the large model to improve the security of the generated text.

[0180] 3. Regarding the problem of poor browsability of the content generated by large models in related technologies, the embodiments of this application can pre-batch produce recommended text content for multiple points of interest of multiple target objects. By combining various post-processing technologies and image-text matching technologies, necessary materials can be effectively added to the generated recommended text to improve the browsability of the generated content.

[0181] Please refer to Figure 4 , which is a method for generating map recommendation information according to an embodiment of this application. This method can be executed by the electronic device 1 shown in Figure 1 . Compared with the foregoing embodiments, the embodiments of this application take the recommendation of surrounding points of interest in map services as an example, aiming to provide users with richer surrounding recommendations and suggestions. The method includes the following steps:

[0182] Step 401: Determine the target object currently queried by the user.

[0183] Step 402: Conduct a surrounding search for the target object to determine multiple points of interest corresponding to the target object.

[0184] Step 403: Score the comment information for each point of interest to determine the recommendation score for each point of interest, and select the to-be-recommended points of interest with the top preset ranking in terms of recommendation scores.

[0185] Step 404: Input the comment information of the to-be-recommended points of interest into the LLM (Large Language Model). The large model LLM compresses and summarizes the comment information of the to-be-recommended points of interest according to the preset prompt words to generate the highlight information of the to-be-recommended points of interest.

[0186] Step 405: Determine the target prompt word corresponding to the target object according to the highlight information of the to-be-recommended points of interest.

[0187] Step 406: Input the target prompt word of the target object into the LLM (large model) so that the large model LLM generates the surrounding recommendation text of the target object according to the target prompt word.

[0188] For the details of each step of the above method, reference can be made to the relevant descriptions of the foregoing embodiments, which will not be elaborated here.

[0189] The solution of the embodiment of the present application can be applied to multiple scenarios, including but not limited to travel guides, scenic spot recommendations, travel planning, and travel services, etc. For example, for a target object, nearby scenic spots, food, shopping, and other related surrounding locations are recommended to the user, and detailed highlight information and evaluations are provided. A variety of AI (Artificial Intelligence) algorithms and technologies are comprehensively used, including natural language processing, machine learning, and deep learning, to seamlessly achieve intelligent travel recommendations and content generation, while ensuring the security and accuracy of the generated text.

[0190] In the travel guide scenario, the embodiment of the present application can help users quickly find nearby scenic spots, food, shopping, etc., and provide detailed highlight information and evaluations. In the scenic spot recommendation scenario, nearby scenic spots can be recommended to the user according to the user's location, and detailed highlight information and evaluations are provided. In the travel planning scenario, it can help users plan the optimal route and the best itinerary arrangement. In the travel service scenario, it can provide users with a more comprehensive travel introduction, etc.

[0191] Please refer to Figure 5 , which is a method for generating recommendation information according to an embodiment of the present application. This method can be executed by the Figure 1 shown electronic device 1 and can be applied to the Figure 2 shown application scenario for generating recommendation information to dynamically provide more accurate recommendation information characterizing the highlights of interest points, meet the diverse needs of users, and enhance the user experience. This embodiment can be executed by the terminal 220. Taking the scenario of online generating recommendation information as an example, the method includes the following steps:

[0192] Step 501: In response to a query instruction for a target object, identify the target category to which the target object belongs.

[0193] Step 502: Determine at least one candidate category associated with the target category.

[0194] Step 503: With the target object as the center point, search for at least one interest point under the candidate category within a preset range.

[0195] Step 504: Obtain the evaluation information and / or introduction information corresponding to each interest point, and input the evaluation information and / or introduction information into a second preset large model, so that the second preset large model generates the first highlight feature of each interest point according to the preset prompt words.

[0196] Step 505: Obtain the evaluation information and / or introduction information corresponding to the target object, and input the evaluation information and / or introduction information into a second preset large model, so that the second preset large model generates the second highlight feature of the target object according to the preset prompt words.

[0197] Step 506: Calculate the recommended scores for each point of interest according to the preset scoring dimensions. For points of interest in the same category, sort multiple points of interest in descending order according to the recommended scores.

[0198] Step 507: Determine at least one point of interest ranked among the top preset number as the point of interest to be recommended under the corresponding category.

[0199] Step 508: Generate a target prompt word based on the first highlight feature of the point of interest to be recommended, the second highlight feature of the target object, and the preset prompt word structure.

[0200] Step 509: Input the target prompt word into the first preset large model so that the first preset large model generates recommendation information about the point of interest according to the target prompt word.

[0201] Step 510: Perform field detection on the recommendation information, correct the error field information in the recommendation information, and output the final recommendation information to the user.

[0202] For the detailed steps of the above method, reference can be made to the relevant descriptions in the foregoing embodiments, which will not be elaborated here.

[0203] Please refer to Figure 6 , which is the generating device 600 for recommendation information according to an embodiment of the present application. This device can be applied to a terminal and can be applied to the generating application scenario of the recommendation information shown in Figure 2 to dynamically provide more accurate recommendation information representing the highlights of the point of interest, meet the diverse needs of users, and enhance the user experience. The device includes: a search module 601, a determination module 602, a construction module 603, and a generation module 604. The functional principles of each module are as follows:

[0204] The search module 601 is configured to search for at least one point of interest corresponding to the target object based on the target object.

[0205] The determination module 602 is configured to determine the first highlight feature of each point of interest through the first description information of each point of interest.

[0206] The construction module 603 is configured to construct a target prompt word based on the first highlight feature.

[0207] The generation module 604 is configured to generate recommendation information corresponding to the target object by using the target prompt word and the first preset large model.

[0208] In an embodiment, the search module 601 is configured to determine at least one associated candidate category based on the category of the target object; and search for at least one point of interest under the candidate category within a preset range with the target object as the center point.

[0209] In one embodiment, a determination module 602 is configured to obtain first description information of each point of interest, where the first description information includes evaluation information and / or introduction information of the point of interest; and input the first description information into a second pre-set large model, so that the second pre-set large model generates first highlight features of each point of interest according to pre-set prompt words.

[0210] In one embodiment, the determination module 602 of the device is further configured to determine second highlight features of a target object based on second description information of the target object; and a construction module 603 is specifically configured to construct a target prompt word based on the first highlight features of the points of interest and the second highlight features of the target object.

[0211] In one embodiment, the construction module 603 is specifically configured to calculate a recommendation score for each point of interest according to a pre-set scoring dimension; sort multiple points of interest in descending order of the recommendation score for points of interest of the same category; determine at least one point of interest ranked in the top pre-set number of places as the point of interest to be recommended under the corresponding category; and generate a target prompt word based on the first highlight features of the points of interest to be recommended, the second highlight features of the target object, and a pre-set prompt word structure, where the pre-set prompt word structure includes a feature slot for the points of interest to be recommended and / or a feature slot for the target object.

[0212] In one embodiment, the pre-set scoring dimension includes one or more of: a highlight feature dimension, a distance dimension, and an information amount dimension; the construction module 603 is specifically configured to, for each point of interest, calculate a feature score of the first highlight features of the point of interest according to the scoring rule of the highlight feature dimension; determine a distance score of the point of interest according to the distance between the point of interest and the target object; determine an information amount score of the point of interest according to the information amount of the first highlight features; and determine the recommendation score of the point of interest according to the feature score, the distance score, the information amount score, and a pre-set weight.

[0213] In one embodiment, it further includes: a detection module, configured to perform field detection on the recommendation information and correct incorrect field information in the recommendation information; and / or an adjustment module, configured to evaluate the output result of the first pre-set large model according to the recommendation information and adjust the first pre-set large model according to the evaluation result.

[0214] For a detailed description of the above-mentioned recommendation information generation device 600, please refer to the description of the relevant method steps in the above-mentioned embodiments. The implementation principles and technical effects are similar, and will not be elaborated here in this embodiment.

[0215] Figure 7 The following is a schematic structural diagram of a cloud device 70 provided by an exemplary embodiment of the present application. The cloud device 70 can be used to run the method provided in any of the above embodiments. As Figure 7 shown, the cloud device 70 may include: a memory 704 and at least one processor 705,Figure 7 Take a processor as an example in

[0216] A memory 704 is used to store computer programs and can be configured to store various other data to support operations on the cloud device 70. The memory 704 can be an Object Storage Service (OSS).

[0217] The memory 704 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0218] A processor 705, coupled to the memory 704, is used to execute the computer program in the memory 704 to implement the solution provided by any of the above method embodiments. The specific functions and achievable technical effects are not elaborated herein.

[0219] Furthermore, as Figure 7 , the cloud device further includes: other components such as a firewall 701, a load balancer 702, a communication component 706, a power supply component 703, etc. Figure 7 Only some components are schematically shown in Figure 7 and it does not mean that the cloud device only includes

[0220] In one embodiment, the above Figure 7The communication component 706 therein is configured to facilitate communication, either wired or wireless, between the device where the communication component 706 is located and other devices. The device where the communication component 706 is located can access a communication standard-based wireless network, such as WiFi, 2G, 3G, 4G, LTE (Long Term Evolution), 5G and other mobile communication networks, or a combination thereof. In an exemplary embodiment, the communication component 706 receives a broadcast signal or broadcast-related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component 706 further includes a Near Field Communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on Radio Frequency Identification (RFID) technology, Infrared Data Association (IrDA) technology, Ultra Wide Band (UWB) technology, Bluetooth (BT) technology and other technologies.

[0221] In one embodiment, the above-mentioned Figure 7 power component 703 supplies power to various components of the device where the power component 703 is located. The power component 703 may include a power management system, one or more power sources, and other components associated with generating, managing, and distributing power for the device where the power component is located.

[0222] The embodiments of the present application also provide a computer-readable storage medium storing computer-executable instructions, and when the processor executes the computer-executable instructions, the method of any of the foregoing embodiments is implemented.

[0223] The embodiments of the present application also provide a computer program product including a computer program, and when the computer program is executed by the processor, the method of any of the foregoing embodiments is implemented.

[0224] In several embodiments provided by the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of modules is only a logical function division, and there may be other division methods in actual implementation. For example, multiple modules can be combined or integrated into another system, or some features can be ignored or not executed.

[0225] The integrated modules implemented in the form of software functional modules can be stored in a computer-readable storage medium. The above-mentioned software functional modules stored in a storage medium include several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) or a processor to execute some steps of the methods of various embodiments of the present application.

[0226] It should be understood that the above-mentioned processor may be a central processing unit (CPU for short), or may also be other general-purpose processors, digital signal processors (DSP for short), application specific integrated circuits (ASIC for short), etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The steps of the method disclosed in combination with the application can be directly embodied as being executed by a hardware processor, or can be executed by a combination of hardware and software modules in the processor. The memory may include a high-speed RAM (Random Access Memory) memory, and may also include a non-volatile storage NVM (Nonvolatile memory for short), such as at least one disk memory, and may also be a USB flash drive, a mobile hard disk, a read-only memory, a magnetic disk, or an optical disc, etc.

[0227] The above-mentioned storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as a static random access memory (SRAM for short), an electrically erasable programmable read-only memory (EEPROM for short), an erasable programmable read-only memory (EPROM for short), a programmable read-only memory (PROM for short), a read-only memory (ROM for short), a magnetic memory, a flash memory, a magnetic disk, or an optical disc. The storage medium can be any available medium that can be accessed by a general-purpose or special-purpose computer.

[0228] An exemplary storage medium is coupled to a processor, enabling the processor to read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can be located in an Application Specific Integrated Circuits (ASIC). Of course, the processor and the storage medium can also exist as discrete components in an electronic device or a master device.

[0229] It should be noted that in this document, the term "comprising", "including" or any other variation thereof is intended to cover non-exclusive inclusion, such that a process, method, clothing or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, clothing or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, clothing or device comprising the element.

[0230] The serial numbers of the embodiments of the present application above are for description only and do not represent the superiority or inferiority of the embodiments.

[0231] Through the description of the above embodiments, those skilled in the art can clearly understand that the above embodiment methods can be implemented by means of software plus a necessary general hardware platform. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation. Based on such an understanding, the technical solution of the present application, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions for causing a terminal device (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods of the various embodiments of the present application.

[0232] In the technical solution of the present application, the processing of collection, storage, use, processing, transmission, provision and disclosure of user data and other information complies with the provisions of relevant laws and regulations and does not violate public order and good customs.

[0233] The above are only the preferred embodiments of the present application and do not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made by using the specification and drawings of the present application, or directly or indirectly applied in other related technical fields, shall be equally included in the patent protection scope of the present application.

Claims

1. A method for generating recommendation information, characterized in that, including: searching for at least one point of interest corresponding to the target object based on the target object; determining the first highlight feature of each point of interest through the first description information of each point of interest; constructing a target prompt based on the first highlight feature; using the target prompt and a first pre-trained large model to generate recommendation information corresponding to the target object.

2. The method according to claim 1, wherein The searching for at least one point of interest corresponding to the target object based on the target object includes: determining at least one associated candidate category based on the category of the target object; searching for at least one of the points of interest under the candidate category within a preset range with the target object as the center point.

3. The method according to claim 1, characterized in that, The determining the first highlight feature of each point of interest through the first description information of each point of interest includes: obtaining the first description information of each point of interest, where the first description information includes evaluation information and / or introduction information of the point of interest; inputting the first description information into a second pre-trained large model so that the second pre-trained large model generates the first highlight feature of each point of interest according to a preset prompt.

4. The method according to claim 1, wherein The method further includes: determining the second highlight feature of the target object based on the second description information of the target object; The constructing a target prompt based on the first highlight feature includes: constructing the target prompt based on the first highlight feature of the point of interest and the second highlight feature of the target object.

5. The method according to claim 4, wherein The constructing the target prompt based on the first highlight feature of the point of interest and the second highlight feature of the target object includes: calculating the recommendation score of each point of interest according to a preset scoring dimension; for points of interest of the same category, sorting multiple points of interest from large to small according to the recommendation score; determining at least one of the points of interest ranked in the top preset number as the to-be-recommended points of interest under the corresponding category; generating the target prompt based on the first highlight feature of the to-be-recommended points of interest, the second highlight feature of the target object, and a preset prompt structure; wherein, the preset prompt structure includes a feature slot about the to-be-recommended points of interest and / or a feature slot about the target object.

6. The method according to claim 5, wherein The preset scoring dimension includes one or more of a highlight feature dimension, a distance dimension, and an information amount dimension; the calculating the recommendation score of each point of interest according to a preset scoring dimension includes: for each point of interest, calculating the feature score of the first highlight feature of the point of interest according to the scoring rule of the highlight feature dimension; determining the distance score of the point of interest according to the distance between the point of interest and the target object; determining the information amount score of the point of interest according to the information amount of the first highlight feature; determining the recommendation score of the point of interest according to the feature score, the distance score, the information amount score, and a preset weight.

7. The method according to any one of claims 1-6, characterized in that, It further includes: performing field detection on the recommendation information and correcting incorrect field information in the recommendation information; and / or, evaluating the output result of the first pre-trained large model according to the recommendation information and adjusting the first pre-trained large model according to the evaluation result.

8. An apparatus for generating recommendation information, characterized in that including: A search module for searching for at least one point of interest corresponding to the target object based on the target object; A determination module for determining the first highlight feature of each point of interest through the first description information of each point of interest; A construction module for constructing a target prompt word based on the first highlight feature; A generation module for generating recommendation information corresponding to the target object by using the target prompt word and a first preset large model.

9. A computer-readable storage medium, characterized in that, Computer-executable instructions are stored in the computer-readable storage medium, and when the processor executes the computer-executable instructions, the method described in any one of claims 1-7 is implemented.

10. A computer program product, characterized in that, It includes a computer program, and when the computer program is executed by the processor, the method described in any one of claims 1-7 is implemented.

Images