A resource on-demand method, system and related apparatus

Abstract

The application discloses a resource on-demand method, system and related device, the method comprises the following steps: obtaining the current demand information input by the current user in the target scene; based on the current demand information, at least one candidate resource is obtained; wherein the candidate resource corresponds to a popularity, and the popularity is obtained based on the number of times that the target resource category included in the historical demand information matches the candidate resource; based on the current demand information and the popularity, a target resource matching the current demand information is determined from the candidate resource, and the target resource is fed back to the current user. Through the above manner, the application can improve the accuracy of resource on-demand.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of natural language processing, in particular to a resource on-demand method, system and related device. BACKGROUND

[0002] With the rapid development of artificial intelligence, the mode of human-computer interaction is constantly evolving and upgrading. Resource on-demand intelligent assistants are ubiquitous in daily life to meet people's entertainment on-demand needs in various scenarios such as smart cars, smart speakers, and smart homes. On-demand resources include music, audio programs, videos, etc. Traditional resource on-demand methods identify resource entity names and entity types in user demand information using semantic understanding models to determine the corresponding target resources. However, in cases where user expression is ambiguous or some content in user demand information is missing, multiple resource types are prone to crosstalk, resulting in poor stability of resource on-demand, thereby affecting the natural and smooth on-demand experience of users. Therefore, how to propose a method for accurately determining on-demand resources based on user demand information has become a problem to be solved. SUMMARY

[0003] The technical problem solved by the present application is to provide a resource on-demand method, system and related device that can improve the accuracy of resource on-demand.

[0004] To solve the above technical problem, one technical solution adopted by the present application is to provide a resource on-demand method, comprising: obtaining current demand information input by a current user in a target scenario; based on the current demand information, obtaining at least one candidate resource; wherein the candidate resource corresponds to a popularity, and the popularity is obtained based on the number of times a target resource category included in historical demand information matches the candidate resource; based on the current demand information and the popularity, determining a target resource that matches the current demand information from the candidate resources, and feeding back the target resource to the current user.

[0005] To solve the above technical problem, another technical solution adopted by the present application is to provide a resource on-demand system, comprising: a first obtaining module for obtaining current demand information input by a current user in a target scenario; a second obtaining module for obtaining at least one candidate resource based on the current demand information; wherein the candidate resource corresponds to a popularity; wherein the popularity is obtained based on the number of times a target resource category included in historical demand information matches the candidate resource; a processing module for determining a target resource that matches the current demand information from the candidate resources based on the current demand information and the popularity, and feeding back the target resource to the current user.

[0006] To solve the above technical problems, another technical solution adopted by the present application is to provide an electronic device, comprising a memory and a processor coupled with each other, the memory stores program instructions, and the processor is configured to execute the program instructions to implement the resource on-demand method mentioned in the above technical solution.

[0007] To solve the above technical problems, another technical solution adopted by the present application is to provide a computer-readable storage medium storing program instructions capable of being executed by a processor, the program instructions being configured to implement the resource on-demand method mentioned in the above technical solution.

[0008] The beneficial effects of the present application are that, unlike the prior art, the resource on-demand method proposed by the present application determines at least one candidate resource after obtaining the current demand information of the user. Moreover, each candidate resource corresponds to a popularity, and the popularity of the candidate resource is determined according to the matching times of the target resource category and the candidate resource in the historical demand information of the current user and the reference user, which is highly accurate. By combining the current demand information of the user and the popularity, the candidate resource with the highest demand fit degree of the user is screened as the target resource, thereby improving the accuracy of user on-demand. BRIEF DESCRIPTION OF DRAWINGS

[0009] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor. Among them:

[0010] Figure 1 is a flowchart of an embodiment of the resource on-demand method of the present application;

[0011] Figure 2 is a flowchart of an embodiment corresponding to step S102;

[0012] Figure 3 is a flowchart of another embodiment corresponding to step S102;

[0013] Figure 4 is a flowchart of another embodiment of the resource on-demand method of the present application;

[0014] Figure 5 is a flowchart of another embodiment corresponding to step S103;

[0015] Figure 6 is a structural diagram of an embodiment of the ambiguity elimination model of the present application;

[0016] Figure 7is a structural schematic diagram of an embodiment of a resource on-demand system of the present application;

[0017] Figure 8 is a structural schematic diagram of an embodiment of an electronic device of the present application;

[0018] Figure 9 is a structural schematic diagram of an embodiment of a computer-readable storage medium of the present application. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments of the present application, and the adaptive combination can be made between different embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0020] The information extraction method proposed in the present application is implemented on the basis of an application, a webpage or at least a smart terminal device integrated with a data processing function on a smart terminal. The smart terminal can be a mobile phone, a tablet computer, a personal computer or a car machine of the current user.

[0021] Please refer to Figure 1 , Figure 1 is a flow schematic diagram of an embodiment of a resource on-demand method of the present application. The method comprises:

[0022] S101: Obtain current demand information input by a current user in a target scenario.

[0023] In an embodiment, in response to the smart terminal having a voice collection function, the current user inputs corresponding demand voice in a voice input manner in the target scenario, so as to serve as the current demand information of the current user.

[0024] In an implementation scenario, the target scenario can be a car machine use scenario or an application use scenario. For example, when the target scenario is a car machine use scenario, the smart terminal corresponds to a car machine system, and the current user can input the current demand information through the car machine system.

[0025] In another embodiment, the current user can also input corresponding demand text in the smart terminal, and take the demand text as the current demand information of the current user.

[0026] S102: Obtain at least one candidate resource based on the current demand information. The candidate resource corresponds to a popularity, and the popularity is obtained based on the number of times of matching a target resource category included in historical demand information with the candidate resource.

[0027] In an embodiment, the obtained current demand information is analyzed to obtain at least one candidate resource matching the current demand information from the resource library.

[0028] Each candidate resource corresponds to a popularity, which is obtained based on the number of times that a target resource category included in the historical demand information matches the candidate resource, the target resource category being an explicit category name in the historical demand information, such as "song" or "novel" or "TV series", etc. In addition, part of the historical demand information can be given by at least one reference user; or, part of the historical demand information can be given by the current user.

[0029] S103: determining a target resource matching the current demand information from the candidate resources based on the current demand information and the popularity, and feeding back the target resource to the current user.

[0030] In an embodiment, the target resource most matching the current demand information of the current user is determined from the candidate resources based on the current demand information and the popularity of the candidate resource corresponding to the current demand information, and the target resource is fed back to the current user.

[0031] The resource on-demand method provided in the present application determines at least one candidate resource corresponding to the current demand information of the user. In addition, each candidate resource corresponds to a popularity, which is determined according to the number of times that a target resource category in the historical demand information of the current user and reference users matches the candidate resource, and has high accuracy. By combining the current demand information of the user and the popularity, the candidate resource with the highest matching degree with the demand of the user is screened as the target resource, thereby improving the accuracy of the on-demand of the user.

[0032] Please refer to Figure 2 , Figure 2 is a flowchart of an embodiment corresponding to step S102. When the current demand information is processed by the intelligent analysis model to obtain the corresponding candidate resource, step S102 specifically includes:

[0033] S201: obtaining intent information and entity information corresponding to the current demand information.

[0034] In an embodiment, a preset template is obtained, and first prompt information is obtained based on the preset template and the current demand information.

[0035] Specifically, the above-mentioned preset template includes a sentence for prompting the intelligent analysis model to perform task analysis, and the current demand information is filled into the corresponding position in the preset template to obtain the first prompt information.

[0036] In a specific embodiment, when the current user expects to play a song and the current demand information is "play #song", a corresponding preset template is obtained, and the current demand information is filled into the preset template to obtain the corresponding first prompt information "play the media resource play assistant, judge whether the 'play #song' given by the current user contains explicit intent information and entity information". It should be noted that "#song" is the name of the song that the current user expects to play in actual application, and the detailed name of the media resource in this application is represented by the corresponding identifier.

[0037] Further, the first prompt information is input into the intelligent analysis model to obtain the intent information and the entity information corresponding to the current demand information.

[0038] Specifically, the first prompt information is input into the intelligent analysis model to enable the intelligent analysis model to analyze the first prompt information and output a corresponding reply, which includes the intent information and the entity information in the current demand information. The entity information is the name of the media resource that the current user expects to play; the intent information includes the play instruction of the current user and the target resource type corresponding to the play instruction, for example, when the current demand information is "play music #song", the play instruction is "play", and the target resource type is "music"; or the intent information can only include the play instruction of the current user, for example, when the current demand information is "play #song", the current demand information only includes the play instruction "play", and the target resource type corresponding to the entity information is not specified.

[0039] In addition, the intelligent analysis model is a large language model with excellent information analysis capability, which generates corresponding prompt information according to the current demand of the current user and the preset template, provides the prompt information to the large language model, and requires the large language model to output the intent information and the entity information contained in the current demand information according to the prompt information. The intelligent analysis model is obtained by fine-tuning a plurality of target training samples, and the target training samples are obtained based on a plurality of initial training samples and a preset template. The target training samples are input into the intelligent analysis model to calculate a corresponding loss value. The loss value is used to adjust part of the parameters in the intelligent analysis model to obtain a fine-tuned intelligent analysis model. The specific fine-tuning process is not described in detail here.

[0040] In a specific application scenario, the large language model can include, but is not limited to, Deep Neural Networks (DNNs), Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), Long Short-Term Memory (LSTM), and generative pre-training Transformer models, and the like. The specific construction and specific deployment of the large language model are not limited herein.

[0041] S202: Based on the intent information and the entity information, at least one candidate resource and its corresponding popularity are obtained.

[0042] In an embodiment, in response to the target resource category being included in the intent information, the target resource category in the intent information is used to search the resource library to obtain the candidate resource matching the entity information.

[0043] In a specific application scenario, the current demand information is "play song #song", that is, it includes the entity information "#song" and the target resource category "song", and the matching candidate resource is searched from the corresponding resource library according to the entity information and the target resource category given by the intelligent analysis model.

[0044] Alternatively, in response to the target resource category not being included in the intent information, at least one candidate resource and its corresponding popularity are obtained based on the entity information.

[0045] In a specific application scenario, when the current demand information of the current user is "play #name", the "#name" is the name of the media resource that the current user expects to play on demand, and in the resource library, the "#name" corresponds to songs, videos, and novels. In response to being unable to accurately know the specific type of media resource that the current user expects to play on demand, the media resources corresponding to the name "#name" in the categories of songs, videos, and novels are all taken as candidate resources.

[0046] In another embodiment, in the process of training the intelligent analysis, the intelligent analysis model is caused to learn a large amount of knowledge information, which includes resource information corresponding to categories such as videos, music, and audio programs, and when the intent information and the entity information in the current demand information of the user are obtained, the intelligent analysis model is caused to search the resource library for candidate resources matching the current demand information according to the learned knowledge information.

[0047] Further, after obtaining at least one candidate resource matching the current demand information through the above steps S201 and S202, the implementation process of step S103 includes:

[0048] In an implementation scenario, in response to the target resource category not being included in the intention information, the candidate resource corresponding to the maximum popularity is taken as the target resource for the multiple candidate resources matching the current demand information, and the target resource is fed back to the user.

[0049] Alternatively, in response to the target resource category not being included in the intention information and the popularities of some candidate resources being relatively close, the candidate resource corresponding to the difference in popularity being within a preset range is fed back to the current user, and the current user is prompted to confirm the fed-back candidate resource. In response to obtaining the confirmation instruction of the current user, the candidate resource corresponding to the confirmation instruction is taken as the target resource.

[0050] In another implementation scenario, in response to the target resource category being included in the intention information, the intelligent analysis model is used to directly search for a unique candidate resource matching the current demand information from the corresponding resource library. On this basis, step S103 need not be performed, and the unique candidate resource is directly taken as the target resource, and the target resource is fed back to the user.

[0051] In yet another implementation scenario, in response to the target resource category being included in the intention information and multiple candidate resources being searched from the corresponding resource library according to the target resource category and the corresponding entity information, step S103 is performed to determine the target resource according to the popularities of different candidate resources. For example, the current demand information of the current user is “playing a song #song”, and the resource library contains multiple song names consistent with the song name in the current demand information, all the corresponding songs are taken as candidate resources, and the target resource is determined according to the corresponding popularities.

[0052] Please refer to Figure 3 , Figure 3 is a flowchart of another implementation of step S102. The candidate resource corresponding to the current demand information can also be obtained based on a semantic analysis model or a preset grammar template, and the specific obtaining manner includes:

[0053] S301: Obtain intention information and entity information corresponding to current demand information.

[0054] In an embodiment, a trained semantic analysis model is obtained, and the current demand information of the current user is input into the semantic analysis model to determine the intent information and the entity information corresponding to the current demand information by performing part-of-speech tagging on different words in the current demand information using the semantic analysis model. The specific structure of the semantic analysis model can refer to an existing semantic analysis neural network structure, and the specific training process is not described in detail here.

[0055] In another embodiment, at least one grammar template is determined in advance, and the obtained current demand information of the current user is compared with the text template to match the content in the current demand information with the slots in the grammar template, thereby determining the intent information and the entity information corresponding to the current demand information. For example, when the grammar template sequentially includes an instruction slot, a type slot, and an entity slot, and the instruction slot is set with the reference word "play", the type slot is set with the reference word "TV series", "song", or sketch, etc., when the current demand information of the current user is "play TV series #name", it is considered that "play" in the current demand information matches the instruction slot in the grammar template, "TV series" in the current demand information matches the type slot in the grammar template, and "#name" in the current demand information matches the entity slot in the grammar template. Based on the above matching relationship, the content in the current demand information that matches the instruction slot and the type slot is taken as the intent information, and the content in the current demand information that matches the entity slot is taken as the entity information.

[0056] In yet another embodiment, the intent information and the entity information corresponding to the current demand information can also be determined by the intelligent analysis model in step S301, and the specific implementation process can refer to the corresponding embodiments described above.

[0057] S302: Based on the intent information and the entity information, at least one candidate resource and its corresponding popularity are obtained.

[0058] In an embodiment, when the intent information is that the current user expects to play the corresponding media resource, at least one candidate resource whose name is consistent with the entity information and the popularity of each candidate resource are obtained from the resource library according to the entity information matched with the intent information, to facilitate subsequent screening.

[0059] Please refer to Figure 4 , Figure 4 is a flowchart of another embodiment of the resource on-demand method of the present application. After obtaining the intent information corresponding to the current demand information through the corresponding embodiments described above, the resource on-demand method proposed in the present application further includes:

[0060] S401: Obtain reference demand information related to the current demand information from the historical demand information input by the current user.

[0061] In an embodiment, the historical demand information of the user input is acquired, and reference demand information at least partially related to the current demand information is acquired therefrom.

[0062] Specifically, the application wake-up instruction is acquired from the historical demand information of the current user, and the wake-up instruction with the closest time stamp to the current time is taken as the reference demand information. For example, in response to the user giving the latest application wake-up instruction “open music player” before the current user gives the current demand information, the “open music player” is taken as the reference demand information.

[0063] In another embodiment, in response to the current demand information being obtained at the current time, and the historical demand information given by the current user containing the corresponding time stamp, the historical demand information with the time stamp within the preset time range from the current time is taken as the reference demand information.

[0064] S402: Based on the reference demand information, the function information and the predicted resource category matching the current demand information are acquired.

[0065] In an embodiment, after the reference demand information is acquired, the matching application information is determined according to the reference demand information, and the function information and the predicted resource category corresponding to the current demand information are acquired according to the application information.

[0066] In an application scenario, the function information is the function supported by the application contained in the current demand information, and the predicted resource category is the resource category with the highest relevance to the application contained in the current demand information.

[0067] In a specific application scenario, when the reference demand information contains “open video player” given by the current user, the corresponding application information is determined to be the video player. In response to the commonly used video player being mainly used for playing video data, but also having the function of playing audio, the corresponding function information acquired includes the video playing function and the audio playing function; and in response to the video playing function having stronger relevance to the video player, the video category is taken as the predicted resource category. Or, when the reference demand information is “open music player”, the corresponding application information is determined to be the music player, the corresponding function information is determined to include the music playing function and the audio book playing function, etc., and the predicted resource category is music. The relevance of the above resource category to the application is related to the click amount of the user, for example, for the music player, most users use the application to play music, and it is determined that the music category has the highest relevance to the music player.

[0068] S403: Based on the function information, the predicted resource category and the intent information, the state information matching the current demand information is acquired.

[0069] In an embodiment, the function information obtained based on the reference demand information, the target resource category in the predicted resource category, and the intention information are combined to obtain the state information.

[0070] In a specific application scenario, an expression template is determined in advance, and the expression template is supplemented according to the function information, the predicted resource category, and the intention information to obtain the state information. Specifically, for the function information, the corresponding position of the expression template contains options such as “music”, “video”, and “audio program”, and based on the specific function contained in the function information, the corresponding position of the above options is marked. For example, the function information contains video playing function and music playing function, and “1” is marked under the corresponding “music” and “video” options, and “0” is marked under the “audio program” option, so that the function information corresponds to the expression “110”. For the predicted resource category, the corresponding position of the expression template also contains options such as “music”, “video”, and “audio program”, and in response to the predicted resource category being “video”, the predicted resource category corresponds to the expression “010”. For the intention information, the corresponding position of the expression template contains options such as “music”, “video”, “audio program”, and “unknown”, and if the intention information does not include an explicit target resource category, the corresponding expression is “0001”.

[0071] In another embodiment, unique labels corresponding to different resource categories are set in advance, such as “music-1”, “video-2”, and “audio program-3”, and when the function information contains video playing function and music playing function, the corresponding expression is “12”.

[0072] In another embodiment, the above state information also includes habit information of the current user. Specifically, when obtaining the above state information, the habit information of the current user is combined, and the habit information is determined based on time. For example, the current user is used to playing music through the car system before 8:00-10:00 in the morning, and when the timestamp corresponding to the current demand information is within the time range, the habit information of the current user is also part of the above state information.

[0073] The above scheme simplifies the function information, the predicted resource category, and the intention information into digital expressions to save the consumption of subsequent computing resources, thereby improving the model analysis efficiency.

[0074] Please refer to Figure 5 and Figure 6 , Figure 5 is a flowchart of step S103 corresponding to another embodiment, Figure 6is a structural schematic diagram of an implementation of the ambiguity resolution model of the present application. After obtaining the intent information and entity information through the steps in the above corresponding implementations, and obtaining the state information matching the current demand information according to the steps S401 to S403, the specific implementation process of step S103 includes:

[0075] S501: input the state information into the first feature extraction network of the ambiguity resolution model to obtain the corresponding state feature.

[0076] In an implementation, the obtained state information is input into the first feature extraction network 11 in the ambiguity resolution model 10, so that the first feature extraction network 11 performs feature extraction on the state information to obtain the corresponding state feature.

[0077] S502: based on the current demand information and the description information of the candidate resource, obtain the second prompt information, input the second prompt information into the second feature extraction network of the ambiguity resolution model to obtain the corresponding prompt feature.

[0078] In an implementation, the description information corresponding to the candidate resource, for example, when the candidate resource is a song, the description information includes the song name, the introduction of the corresponding singer, the song release time and the album to which it belongs, etc. The current demand information and the description information of each candidate resource are combined to obtain the second prompt information, which is input into the second feature extraction network 12 in the ambiguity resolution model 10, so that the second feature extraction network 12 performs feature extraction on the second prompt information to obtain the corresponding prompt feature.

[0079] It should be noted that in actual application, the order of obtaining the state feature and the prompt feature can be adaptively adjusted, for example, step S502 is performed first to obtain the prompt feature, and then step S501 is performed to obtain the state feature; or the prompt feature and the state feature are obtained at the same time.

[0080] S503: input the state feature and the prompt feature into the fusion network of the ambiguity resolution model to obtain the fusion feature.

[0081] In an embodiment, the obtained state feature and prompt feature are input into the fusion network 13 in the ambiguity resolution model 10, so that the fusion network 13 analyzes the state feature and the prompt feature, and determines the attention weight corresponding to the state feature and the prompt feature respectively. Based on the state feature and the first attention weight corresponding thereto, and the prompt feature and the second attention weight corresponding thereto, the fusion feature is obtained. The first attention weight and the second attention weight are determined to enable the ambiguity resolution model 10 to actively pay attention to information that is more conducive to candidate resource judgment, thereby improving the accuracy of subsequent target resource determination. The specific structure of the fusion network 13 can refer to MoE (Mixture of Experts, mixed expert model). In addition, the specific calculation formula of the fusion feature is as follows:

[0082] h out =w i *h text +(1-w i )*h system

[0083] wherein h text represents the prompt feature, w i represents the first attention weight corresponding to the prompt feature, h system represents the state feature, (1-w i ) represents the second attention weight corresponding to the state feature, and h out represents the fusion feature.

[0084] In addition, the ambiguity resolution model is obtained by training a plurality of training samples, and the specific training process will not be described in detail herein.

[0085] S504: Based on the fusion feature and the popularity corresponding to the candidate resource, obtain the confidence score corresponding to each candidate resource.

[0086] In an embodiment, the obtained fusion feature and the popularity corresponding to each candidate resource are input into the full connection layer 14 in the ambiguity resolution model 10 to obtain the confidence score corresponding to each candidate resource. The higher the confidence score is, the higher the probability that the corresponding candidate resource is the target resource is.

[0087] In an implementation scenario, the full connection layer 14 preferentially analyzes the fusion feature and outputs the reference score corresponding to each candidate resource corresponding to the fusion feature. For each candidate resource, the corresponding reference score and the popularity are weighted and summed to obtain the confidence score corresponding to the candidate resource.

[0088] S505: Based on the confidence score, determine the target resource from all candidate resources.

[0089] In an embodiment, in response to obtaining the confidence score corresponding to each candidate resource, the candidate resource corresponding to the maximum confidence score is taken as the target resource.

[0090] In another embodiment, in response to the existence of multiple confidence scores greater than the preset threshold value in the confidence scores corresponding to all candidate resources, prompt content for prompting the current user to make a determination is generated, and the candidate resource with the part of the confidence score greater than the preset threshold value and the prompt content are fed back to the current user.

[0091] Further, in response to obtaining a selection instruction triggered by the current user based on the prompt content, the candidate resource corresponding to the selection instruction is taken as the target resource.

[0092] Alternatively, in response to the current user giving the corresponding selection instruction within a preset time range, the candidate resource corresponding to the maximum confidence score is taken as the target resource.

[0093] In an embodiment, the popularity of the candidate resource mentioned in any of the above embodiments is obtained based on the number of times of matching the target resource category and the candidate resource included in the historical demand information. Wherein, in response to any historical demand information including entity information and target resource category, and the entity information in the historical demand information being consistent with the name of the candidate resource and the target resource category being consistent with the category corresponding to the candidate resource, it is determined that the historical demand information matches the candidate resource. Therefore, for any resource in the database, the determination process of the corresponding popularity includes:

[0094] The historical demand information including entity information and target resource category is taken as statistical information, and for the current candidate resource, the ratio of the statistical information matching the candidate resource to the number of all statistical information including only the entity information corresponding to the candidate resource is taken as the popularity of the candidate resource. Wherein, taking the music resource with the name "#name" as an example, the specific calculation formula of the popularity is as follows:

[0095] P(d, #name) = N d (#name) / ∑ s∈i N s (#name)

[0096] Wherein, d represents the music resource category, P(f, #name) represents the popularity of the music resource with the name "#name", N d (#name) represents the number of statistical information including the music resource with the name "#name", i represents the set corresponding to all resource categories, N s (#name) represents the number of statistical information including the name "#name".

[0097] In addition, it should be noted that the historical demand information used to determine the popularity can be historical demand information obtained in the current scenario. For example, in response to the current demand information being given by the current user on the car machine system, the popularity of different candidate resources is determined based on historical demand information given by multiple reference users on the corresponding car machine system. This way makes the determined popularity of different candidate resources more suitable for the current scenario, thereby improving the stability of user resource on-demand.

[0098] In a specific application scenario, 100 pieces of historical demand information given by multiple reference users are obtained, and the 100 pieces of historical demand information all contain entity information and corresponding target resource categories. Among them, the number of historical demand information containing entity information "#name" is 60, and the number of historical demand information containing entity information "#name" and corresponding target resource category "music" is 50, and the corresponding candidate resource is determined. The popularity of the resource is 0.83.

[0099] In another embodiment, the reference popularity of the candidate resource can also be determined only according to the historical demand information of the current user, that is, the past on-demand information of the current user. For example, when the current scenario is a car machine system use scenario, and the current user has on-demanded entity information "#name" and target resource type "music" in the car machine system for 5 times in the past, and has not on-demanded entity information "#name" and target resource type "resources different from music", then the corresponding reference popularity is 1.

[0100] Further, for the same candidate resource, the reference popularity obtained based on the historical demand information of the current user and the popularity determined based on multiple reference users are weighted and summed to obtain the final popularity of each candidate resource.

[0101] Please refer to Figure 7 , Figure 7 is a structural schematic diagram of an embodiment of a resource on-demand system of the present application. The resource on-demand system includes a first acquisition module 20, a second acquisition module 30, and a processing module 40 which are coupled to each other.

[0102] Specifically, the first acquisition module 20 is configured to acquire current demand information input by a current user in a target scenario.

[0103] The second acquisition module 30 is configured to acquire at least one candidate resource based on the current demand information; wherein the candidate resource corresponds to a popularity; wherein the popularity is obtained based on the number of times that the target resource category included in the historical demand information matches the candidate resource.

[0104] The processing module 40 is configured to determine a target resource matching the current demand information from the candidate resources based on the current demand information and the popularity, and feed back the target resource to the current user.

[0105] In an embodiment, the second obtaining module 30 obtains at least one candidate resource based on the current demand information, including: obtaining intent information and entity information corresponding to the current demand information; and obtaining at least one candidate resource and its corresponding popularity based on the intent information and the entity information.

[0106] In an embodiment, the second obtaining module 30 obtains intent information and entity information corresponding to the current demand information, including: obtaining a preset template, and obtaining first prompt information based on the preset template and the current demand information; and inputting the first prompt information into an intelligent analysis model to obtain the intent information and the entity information corresponding to the current demand information; wherein the intelligent analysis model is obtained by fine-tuning a plurality of target training samples, and the target training samples are obtained based on a plurality of initial training samples and the preset template.

[0107] In an embodiment, the second obtaining module 30 obtains at least one candidate resource and its corresponding popularity based on the intent information and the entity information, including: in response to the target resource category being included in the intent information, obtaining a candidate resource matching the entity information based on the target resource category; and in response to the target resource category not being included in the intent information, obtaining at least one candidate resource and its corresponding popularity based on the entity information.

[0108] In an embodiment, in response to the target resource category not being included in the intent information, the processing module 40 determines a target resource matching the current demand information from the candidate resources based on the current demand information and the popularity, including: taking the candidate resource corresponding to the maximum popularity as the target resource.

[0109] In an embodiment, please refer to Figure 7 The resource on-demand system provided in the present application further includes a state information obtaining module 50 coupled with the processing module 40, the state information obtaining module 50 is configured to obtain historical demand information related to the current demand information input by the current user, obtain function information matching the current demand information based on the historical demand information; and obtain state information matching the current demand information based on the function information and the intent information.

[0110] In an embodiment, in response to the candidate resource corresponding to the description information, the processing module 40 determines the target resource matching the current demand information from the candidate resource based on the current demand information and the popularity, including: inputting the state information to the first feature extraction network of the disambiguation model to obtain the corresponding state feature; and, based on the current demand information and the description information of the candidate resource, obtaining the second prompt information, inputting the second prompt information to the second feature extraction network of the disambiguation model to obtain the corresponding prompt feature; inputting the state feature and the prompt feature to the fusion network of the disambiguation model to obtain the fusion feature; based on the fusion feature and the popularity corresponding to the candidate resource, obtaining the confidence score corresponding to each candidate resource; based on the confidence score, determining the target resource from all candidate resources.

[0111] In an embodiment, please refer to Figure 7 The resource on-demand system proposed in the present application also includes a confirmation module 60 coupled with the processing module 40, which is used to take the candidate resource corresponding to the maximum confidence score as the target resource. Alternatively, in response to the existence of multiple confidence scores greater than a preset threshold, a prompt content for prompting the current user to confirm is generated, and the candidate resource corresponding to the confidence score greater than the preset threshold and the prompt content are fed back to the current user; in response to obtaining the selection instruction triggered by the current user based on the prompt content, the candidate resource corresponding to the selection instruction is taken as the target resource.

[0112] Please refer to Figure 8 , Figure 8 is a structural schematic diagram of an embodiment of an electronic device of the present application. The electronic device includes a memory 70 and a processor 80 coupled with each other. The memory 70 stores program instructions, and the processor 80 is used to execute the program instructions to realize the resource on-demand method mentioned in any of the above embodiments. Specifically, the electronic device includes but is not limited to a desktop computer, a notebook computer, a tablet computer, a server, etc., which are not limited herein. In addition, the processor 80 can also be referred to as a CPU (Center Processing Unit). The processor 80 can be an integrated circuit chip with signal processing capability. The processor 80 can also be a general processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field-Programmable Gate Array) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. The general processor can be a microprocessor or the processor can also be any conventional processor. In addition, the processor 80 can be realized by integrated circuit chips together.

[0113] Please refer to Figure 9 , Figure 9 is a structural schematic diagram of an embodiment of the computer readable storage medium of the present application, the computer readable storage medium 90 stores program instructions 95 capable of being executed by a processor, and the program instructions 95 are executed by the processor to implement the resource on-demand method mentioned in any of the above embodiments.

[0114] In several embodiments provided in the present application, it should be understood that the disclosed method and device can be implemented in other ways. For example, the above-described device embodiment is only schematic, for example, the division of the module or unit is only a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interface, device or unit, and can be electrical, mechanical or other forms.

[0115] The unit described as a separate component can be or can not be physically separated, and the component shown as a unit can be or can not be a physical unit, that is, it can be located in one place, or can be distributed on a plurality of network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the present embodiment scheme.

[0116] In addition, the function units in each embodiment of the present application can be integrated in one processing unit, or each unit can be physically present alone, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software function unit.

[0117] If the integrated unit is realized in the form of a software function unit and sold or used as an independent product, it can be stored in a computer readable storage medium. Based on such understanding, the technical scheme of the present application or the essential part or all or part of the technical scheme that contributes to the prior art can be embodied in the form of a software product. The computer software product is stored in a storage medium, including a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to execute all or part of the steps of the method of each embodiment of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0118] The above merely provides the implementation of the present application, and does not limit the patent scope of the present application, and any equivalent structure or equivalent process transformation made by using the content of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A resource on-demand method, characterized in that, include: Obtain the current user's current requirement information in the target scenario; Based on the current demand information, at least one candidate resource is obtained; wherein, the candidate resource corresponds to a popularity, and the popularity is obtained based on the number of times the target resource category included in the historical demand information matches the candidate resource; Based on the current demand information and the popularity, a target resource matching the current demand information is determined from the candidate resources, and the target resource is fed back to the current user. The step of obtaining at least one candidate resource based on the current demand information includes: obtaining intent information and entity information corresponding to the current demand information; and obtaining at least one candidate resource and its corresponding popularity based on the intent information and the entity information. In response to the candidate resources having corresponding descriptive information, and based on the current demand information and the popularity, a target resource matching the current demand information is determined from the candidate resources, including: inputting state information into a first feature extraction network of a disambiguation model to obtain corresponding state features; wherein the state information is determined based on functional information matching the current demand information, predicted resource category, and intent information; and, based on the current demand information and the descriptive information of the candidate resources, obtaining second prompt information, inputting the second prompt information into a second feature extraction network of the disambiguation model to obtain corresponding prompt features; inputting the state features and the prompt features into a fusion network of the disambiguation model to obtain fused features; obtaining a confidence score corresponding to each candidate resource based on the fused features and the popularity corresponding to the candidate resources; and determining the target resource from all the candidate resources based on the confidence score.

2. The method according to claim 1, characterized in that, The step of obtaining the intent information and entity information corresponding to the current demand information includes: Obtain a preset template, and based on the preset template and the current requirement information, obtain a first prompt message; The first prompt information is input into the intelligent analysis model to obtain the intent information and entity information corresponding to the current demand information; wherein, the intelligent analysis model is obtained by fine-tuning multiple target training samples, and the target training samples are obtained based on multiple initial training samples and preset templates.

3. The method according to claim 2, characterized in that, The step of obtaining at least one candidate resource and its corresponding popularity based on the intent information and the entity information includes: In response to the intent information including a target resource category, the candidate resource matching the entity information is obtained based on the target resource category; In response to the absence of the target resource category in the intent information, at least one candidate resource and its corresponding popularity are obtained based on the entity information.

4. The method according to claim 3, characterized in that, In response to the fact that the target resource category is not included in the intent information, the step of determining the target resource matching the current demand information from the candidate resources based on the current demand information and the popularity includes: The candidate resource with the highest popularity is selected as the target resource.

5. The method according to claim 1, characterized in that, The step of determining the target resource from all candidate resources based on the confidence score includes: The candidate resource corresponding to the highest confidence score is taken as the target resource; Alternatively, in response to the existence of multiple confidence scores greater than a preset threshold, a prompt message is generated to prompt the current user to confirm, and the candidate resources with corresponding confidence scores greater than the preset threshold and the prompt message are fed back to the current user; In response to receiving a selection instruction triggered by the current user based on the prompt content, the candidate resource corresponding to the selection instruction is taken as the target resource.

6. A resource on-demand system, characterized in that, include: The first acquisition module is used to acquire the current user's current requirement information in the target scenario; The second acquisition module is used to acquire at least one candidate resource based on the current demand information; wherein the candidate resource corresponds to a popularity; wherein the popularity is obtained based on the number of times the target resource category included in the historical demand information matches the candidate resource; The processing module is used to determine the target resource that matches the current demand information from the candidate resources based on the current demand information and the popularity, and to feed back the target resource to the current user; The step of obtaining at least one candidate resource based on the current demand information includes: obtaining intent information and entity information corresponding to the current demand information; and obtaining at least one candidate resource and its corresponding popularity based on the intent information and the entity information. In response to the candidate resources having corresponding descriptive information, and based on the current demand information and the popularity, determining the target resource matching the current demand information from the candidate resources includes: inputting state information into a first feature extraction network of a disambiguation model to obtain corresponding state features; wherein the state information is determined based on functional information matching the current demand information, predicted resource category, and intent information; and, based on the current demand information and the descriptive information of the candidate resources, obtaining second prompt information, inputting the second prompt information into a second feature extraction network of the disambiguation model to obtain corresponding prompt features; inputting the state features and the prompt features into a fusion network of the disambiguation model to obtain fused features; obtaining a confidence score corresponding to each candidate resource based on the fused features and the popularity corresponding to the candidate resources; and determining the target resource from all the candidate resources based on the confidence score.

7. An electronic device, characterized in that, The method includes a memory and a processor coupled to each other, wherein the memory stores program instructions and the processor executes the program instructions to implement the resource on-demand method according to any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, The system stores program instructions that can be executed by a processor, the program instructions being used to implement the resource on-demand method according to any one of claims 1-5.

Images