Intention information identification method and device, intention information training method and device, storage medium and equipment

Abstract

The invention provides an intention information identification method and device, a training method and device, a storage medium and equipment, and relates to the field of artificial intelligence and the field of digital marketing. The method comprises the following steps: extracting text features of a description text to obtain target description text features; according to a first similarity between the target description text feature and an initial intention text feature, a first recognition result for the initial intention information is determined, and the initial intention text feature represents the initial intention information; a candidate intention corpus matched with the description text is determined, and the candidate intention corpus is associated with an intention association weight corresponding to the initial intention information; determining a second recognition result for the initial intention information according to the intention association weight and a second similarity between the description text and the candidate intention corpus; and obtaining target intention information according to the first recognition result and the second recognition result.

Description

Technical Field

[0001] The present disclosure relates to the fields of artificial intelligence and digital marketing, and more specifically, to an intent information recognition method, training method, device, storage medium, equipment and program product. Background Art

[0002] With the rapid development of Internet technology, relevant e-commerce application platforms can meet the diverse needs of users by pushing product links and intelligent customer service information that meet user intentions. For example, data analysis can be performed on the description text entered by the user to identify the user's intention, and corresponding marketing resources can be pushed based on the identified intention, or intelligent reply information related to the intention can be sent.

[0003] In the process of realizing the concept of the present disclosure, the inventors found that there are at least the following problems in the related art: the accuracy of user intention recognition in the related art is low, and it is difficult to push relevant information that meets the user's intention by accurately identifying the user's intention. Summary of the invention

[0004] In view of this, the present disclosure provides an intent information recognition method, training method, apparatus, storage medium, device and program product.

[0005] One aspect of the present disclosure provides a method for identifying intent information, comprising:

[0006] Extract the text features of the description text to obtain the target description text features;

[0007] Determining a first recognition result for the initial intent information according to a first similarity between the target description text feature and the initial intent text feature, wherein the initial intent text feature represents the initial intent information;

[0008] Determine a candidate intent corpus matching the description text, wherein the candidate intent corpus is associated with an intent association weight corresponding to the initial intent information;

[0009] Determine a second recognition result for the initial intent information according to the intent association weight and a second similarity between the description text and the candidate intent corpus;

[0010] According to the above-mentioned first recognition result and the above-mentioned second recognition result, the target intention information is obtained.

[0011] According to an embodiment of the present disclosure,

[0012] According to an embodiment of the present disclosure, the above-mentioned intention information recognition method further includes:

[0013] Obtaining product display information and product operation information corresponding to the initial intent corpus, wherein the initial intent corpus is extracted from the user's historical query information;

[0014] Determining initial intention information associated with the target commodity category according to the target commodity category represented by the commodity operation information; and

[0015] According to the product display information and the product operation information, an intention association weight between the initial intention corpus and the initial intention information is determined.

[0016] According to an embodiment of the present disclosure, the candidate intent corpus that is determined to match the description text includes:

[0017] Extracting the intended keywords from the above description text; and

[0018] Determine a candidate intent corpus matching the intent keyword from an initial intent corpus set including the initial intent corpus.

[0019] According to an embodiment of the present disclosure, the above-mentioned extraction of text features of the description text includes:

[0020] Extract description keywords from the above description text;

[0021] Extract the text features of the above-mentioned description keywords to obtain the text features of the description keywords;

[0022] Extracting the text features of the description text to obtain a first description text feature; and

[0023] The first description text feature and the description keyword text feature are integrated.

[0024] According to an embodiment of the present disclosure, the above-mentioned description keyword text features include multiple;

[0025] The above-mentioned fusion of the above-mentioned first description text feature and the above-mentioned description keyword text feature includes:

[0026] Merging the plurality of description keyword text features with the first description text feature to obtain a second description text feature corresponding to the description keyword text feature; and

[0027] Based on the attention mechanism, the second description text features corresponding to the plurality of the description keyword text features are fused.

[0028] According to an embodiment of the present disclosure, the first recognition result includes a first recognition value, and the second recognition result includes a second recognition value;

[0029] The target intention information obtained according to the first recognition result and the second recognition result includes:

[0030] Determine a first target recognition value according to the first recognition value and the first recognition weight;

[0031] Determine a second target recognition value according to the second recognition value and the second recognition weight;

[0032] Determining a third target identification value based on the first target identification value and the second target identification value; and

[0033] When the third target recognition value is greater than a preset recognition threshold, the initial intention information is recognized as target intention information.

[0034] According to an embodiment of the present disclosure, the above-mentioned intention information recognition method further includes:

[0035] Push recommended resource information corresponding to the above target intent information to the target client corresponding to the above description text.

[0036] Another aspect of the present disclosure also provides a method for training a deep learning model, comprising:

[0037] Acquire a training sample, wherein the training sample includes a sample intent corpus and a sample label, and the sample label includes sample intent information and a sample intent association weight corresponding to the sample intent information;

[0038] Input the sample intent corpus into the first text feature extraction layer of the initial deep learning model, and output the predicted intent corpus text features;

[0039] Input the sample intent information into the second text feature extraction layer of the initial deep learning model, and output the sample intent text features;

[0040] The initial deep learning model is trained using the predicted intent corpus text features, the sample intent text features, and the sample intent association weights to obtain a trained deep learning model.

[0041] According to an embodiment of the present disclosure, the above-mentioned training of the above-mentioned initial deep learning model using the above-mentioned predicted intent corpus text features, the above-mentioned sample intent text features and the above-mentioned sample intent association weights includes:

[0042] Update the initial loss function according to the above sample intention association weights to obtain the target loss function;

[0043] Input the above-mentioned prediction intent corpus text features and the above-mentioned sample intent text features into the above-mentioned target loss function to obtain a target loss value;

[0044] The parameters of the initial deep learning model are adjusted based on the target loss value to obtain the trained deep learning model.

[0045] According to an embodiment of the present disclosure, updating the initial loss function according to the sample intention association weights includes:

[0046] Update the sample intention association weight based on the preset rules to obtain the sample target weight; and

[0047] The above initial loss function is updated using the above sample target weights.

[0048] Another aspect of the present disclosure further provides an intention information recognition device, comprising:

[0049] A text feature extraction module is used to extract text features of the description text to obtain target description text features;

[0050] A first recognition result determination module, configured to determine a first recognition result for the initial intent information according to a first similarity between the target description text feature and the initial intent text feature, wherein the initial intent text feature represents the initial intent information;

[0051] A candidate intent corpus determination module is used to determine a candidate intent corpus that matches the description text, wherein the candidate intent corpus is associated with an intent association weight corresponding to the initial intent information;

[0052] A second recognition result determination module, configured to determine a second recognition result for the initial intent information according to the intent association weight and a second similarity between the description text and the candidate intent corpus;

[0053] The target intention information determination module is used to obtain the target intention information based on the above-mentioned first recognition result and the above-mentioned second recognition result.

[0054] Another aspect of the present disclosure further provides a training device for a deep learning model, comprising:

[0055] A training sample acquisition module, used to acquire training samples, wherein the training samples include sample intent corpus and sample labels, and the sample labels include sample intent information and sample intent association weights corresponding to the sample intent information;

[0056] A module for determining text features of a predicted intent corpus, configured to input the sample intent corpus into the first text feature extraction layer of the initial deep learning model, and output text features of the predicted intent corpus;

[0057] A sample intent text feature determination module, used to input the sample intent information into the second text feature extraction layer of the initial deep learning model, and output the sample intent text feature;

[0058] The training module is used to train the initial deep learning model using the predicted intent corpus text features, the sample intent text features and the sample intent association weights to obtain a trained deep learning model.

[0059] Another aspect of the present disclosure provides an electronic device, comprising:

[0060] one or more processors;

[0061] a memory for storing one or more programs,

[0062] When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described above.

[0063] Another aspect of the present disclosure provides a computer-readable storage medium storing computer-executable instructions, which are used to implement the above method when executed.

[0064] Another aspect of the present disclosure provides a computer program product, which includes computer executable instructions, and when the instructions are executed, are used to implement the method as described above.

[0065] According to the embodiments of the present disclosure, because a first similarity is determined between the target description text features and the initial intent text features representing the initial intent information, and a second similarity is determined between the candidate intent corpus matching the description text and the description text, the similarity between the description text and the initial intent information can be comprehensively considered from multiple dimensions of the intent information and the intent corpus, and the target intent information can be identified based on the first recognition result and the second recognition result. This can solve the technical problem of low recognition accuracy of the description text for the target intent information, reduce the probability of intent recognition errors, and achieve the technical effect of improving the accuracy of related resource recommendations. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] The above and other objects, features and advantages of the present disclosure will become more apparent through the following description of the embodiments of the present disclosure with reference to the accompanying drawings, in which:

[0067] Figure 1 An exemplary system architecture to which the intent recognition method and device according to an embodiment of the present disclosure can be applied is schematically shown;

[0068] Figure 2 A flowchart of a method for identifying intention information according to an embodiment of the present disclosure is schematically shown;

[0069] Figure 3 A flowchart of extracting text features of a description text according to an embodiment of the present disclosure is schematically shown;

[0070] Figure 4 The application scenario diagram of the intention information recognition method according to the embodiment of the present disclosure is schematically shown;

[0071] Figure 5 The application scenario diagram of the intention information recognition method according to the embodiment of the present disclosure is schematically shown;

[0072] Figure 6 A block diagram schematically shows an apparatus for identifying intention information according to an embodiment of the present disclosure;

[0073] Figure 7 A block diagram schematically illustrates a training device for a deep learning model according to an embodiment of the present disclosure; and

[0074] Figure 8 A block diagram of an electronic device suitable for implementing a method for identifying intent information and a method for training a deep learning model according to an embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION

[0075] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present disclosure. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0076] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise", "include", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0077] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification, and should not be interpreted in an idealized or overly rigid manner.

[0078] When using expressions such as "at least one of A, B, and C, etc.", they should generally be interpreted according to the meaning of the expression commonly understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).

[0079] In the embodiments of the present disclosure, the collection, updating, analysis, processing, use, transmission, provision, disclosure, storage, etc. of the data involved (for example, including but not limited to user personal information) are in compliance with the provisions of relevant laws and regulations, are used for legitimate purposes, and do not violate public order and good morals. In particular, necessary measures are taken for user personal information to prevent illegal access to user personal information data and maintain the security of user personal information, network security, and national security.

[0080] In the embodiments of the present disclosure, the user's authorization or consent is obtained before obtaining or collecting the user's personal information.

[0081] With the rapid development of Internet technology, relevant e-commerce application platforms can meet the diverse needs of users by pushing product connections and intelligent customer service information that meet the user's intentions. Usually, data analysis can be performed on the description text entered by the user to identify the user's intention. However, the existing intent recognition algorithm needs to be trained based on a manually annotated dialogue corpus to understand the user's intention, which places certain requirements on the scale and quality of the dialogue corpus. As a result, it is difficult to train an intent recognition model with accurate intent prediction when the corpus quality is low or the amount of data is small. At the same time, it is difficult to correct intent recognition errors in real time when the intent recognition model is trained offline, resulting in poor training effect of the intent recognition model and low recognition accuracy.

[0082] The embodiments of the present disclosure provide an intent information recognition method, training method, device, storage medium, equipment and program product. The intent information recognition method includes: extracting text features of a description text to obtain target description text features; determining a first recognition result for the initial intent information according to a first similarity between the target description text features and the initial intent text features, wherein the initial intent text features characterize the initial intent information; determining a candidate intent corpus matching the description text, wherein the candidate intent corpus is associated with an intent association weight corresponding to the initial intent information; determining a second recognition result for the initial intent information according to the intent association weight and a second similarity between the description text and the candidate intent corpus; and obtaining the target intent information according to the first recognition result and the second recognition result.

[0083] According to an embodiment of the present disclosure, by determining a first similarity between target description text features and initial intent text features representing initial intent information, and determining a second similarity between a candidate intent corpus matching the description text and the description text, the similarity between the description text and the initial intent information can be comprehensively considered from multiple dimensions of intent information and intent corpus, and then the target intent information can be identified based on the first recognition result and the second recognition result, thereby improving the recognition accuracy of the description text, reducing the probability of intent recognition errors, and achieving the technical effect of improving the accuracy of related resource recommendations.

[0084] Figure 1 The following schematically illustrates an exemplary system architecture to which the intent recognition method and device according to an embodiment of the present disclosure can be applied. It should be noted that: Figure 1 What is shown is merely an example of a system architecture to which the embodiments of the present disclosure can be applied, in order to help those skilled in the art understand the technical content of the present disclosure, but it does not mean that the embodiments of the present disclosure cannot be used in other devices, systems, environments or scenarios.

[0085] like Figure 1 As shown, the system architecture 100 according to this embodiment may include terminal devices 101, 102, 103, a network 104 and a server 105. The network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired and / or wireless communication links, etc.

[0086] Users can use terminal devices 101, 102, 103 to interact with server 105 through network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients and / or social platform software, etc. (only as examples).

[0087] The terminal devices 101 , 102 , and 103 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, and desktop computers.

[0088] The server 105 may be a server that provides various services, such as a background management server (only an example) that provides support for websites browsed by users using the terminal devices 101, 102, and 103. The background management server may analyze and process the received data such as user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal device.

[0089] It should be noted that the intent information identification method provided in the embodiment of the present disclosure can generally be executed by the server 105. Accordingly, the intent information identification device provided in the embodiment of the present disclosure can generally be set in the server 105. The intent information identification method provided in the embodiment of the present disclosure can also be executed by a server or server cluster that is different from the server 105 and can communicate with the terminal devices 101, 102, 103 and / or the server 105. Correspondingly, the intent information identification device provided in the embodiment of the present disclosure can also be set in a server or server cluster that is different from the server 105 and can communicate with the terminal devices 101, 102, 103 and / or the server 105. Alternatively, the intent information identification method provided in the embodiment of the present disclosure can also be executed by the terminal devices 101, 102, or 103, or can also be executed by other terminal devices different from the terminal devices 101, 102, or 103. Correspondingly, the intention information identification device provided in the embodiments of the present disclosure may also be set in the terminal device 101, 102, or 103, or in other terminal devices different from the terminal device 101, 102, or 103.

[0090] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0091] Figure 2 The flowchart of the method for identifying intention information according to an embodiment of the present disclosure is schematically shown.

[0092] like Figure 2 As shown, the intention information recognition method includes operations S210 to S250.

[0093] In operation S210, text features of the description text are extracted to obtain target description text features.

[0094] According to an embodiment of the present disclosure, the description text may be text information used to describe user needs, such as a commodity consultation text, a query text, a communication text, etc. input by a user. The description text may be formed based on the ideas or needs generated by the user in a short period of time, so the description text may be colloquial and have low grammatical norms.

[0095] According to the embodiments of the present disclosure, the target description text feature may be feature information used to characterize the semantic information of the description text. The target description text feature may be extracted based on a neural network algorithm, for example, the text feature of the description text may be extracted based on an attention network model, but is not limited thereto. The text feature of the description text may also be extracted based on a recurrent neural network algorithm. The embodiments of the present disclosure do not limit the specific algorithm type for extracting the text feature of the description text, and those skilled in the art may select one according to actual needs.

[0096] In operation S220, a first recognition result for the initial intent information is determined according to a first similarity between the target description text feature and the initial intent text feature, wherein the initial intent text feature represents the initial intent information.

[0097] According to an embodiment of the present disclosure, the initial intention information may include text information representing the user's commodity demand intention. Exemplarily, the initial intention information may be, for example, "recommendation of household daily necessities commodities".

[0098] According to the embodiments of the present disclosure, the initial intent text feature may be feature information that can characterize the semantic information of the initial intent information. The initial intent text feature may be obtained by encoding the initial intent information, but is not limited thereto. The initial intent information may also be processed based on a neural network model to obtain the initial intent text feature. The embodiments of the present disclosure do not limit the specific method for obtaining the initial intent text feature, and those skilled in the art may make a selection based on actual needs.

[0099] According to an embodiment of the present disclosure, the initial intent text features and the target description text features can be processed based on a similarity algorithm to obtain a first similarity. The similarity algorithm can be, for example, a cosine similarity algorithm to obtain the first similarity, but is not limited thereto, and can also be calculated based on other types of algorithms to obtain the first similarity, which is not limited in the embodiment of the present disclosure.

[0100] According to an embodiment of the present disclosure, the first recognition result may be an evaluation result characterizing the degree of semantic similarity between the initial intent information and the description text. For example, the first recognition result may be characterized based on a recognition score.

[0101] In operation S230 , candidate intent corpora matching the description text are determined, wherein the candidate intent corpora are associated with intent association weights corresponding to the initial intent information.

[0102] According to an embodiment of the present disclosure, the candidate intent corpus may be a text corpus such as commodity consultation text, query text, communication text, etc. collected in a historical period. The intent association weight may characterize the similarity between the candidate intent corpus and the initial intent information.

[0103] For example, the candidate intent corpus can be: "I want to buy laundry supplies", the intent association weight between the candidate intent corpus and the initial intent information "recommendations for home appliances" can be 0.3, and the intent association weight between the candidate intent corpus and the initial intent information "recommendations for household daily necessities" can be 0.7.

[0104] It should be understood that the candidate intent corpus may be associated with one or more initial intent information, and correspondingly, the initial intent information may also be associated with one or more candidate intent corpora.

[0105] According to the embodiments of the present disclosure, the candidate intent corpus that matches the description text can be determined by text matching. However, it is not limited to this. The similarity between the description text and the initial intent corpus can also be calculated based on a similarity algorithm, and then the candidate intent corpus that matches the description text can be determined based on the similarity. The embodiments of the present disclosure do not limit the specific method for determining the candidate intent corpus, and those skilled in the art can make a choice according to actual needs.

[0106] In operation S240 , a second recognition result for the initial intent information is determined according to the intent association weight and a second similarity between the description text and the candidate intent corpus.

[0107] According to an embodiment of the present disclosure, the description text and the candidate intent corpus may be processed based on a similarity algorithm to obtain a second similarity. The similarity algorithm may be, for example, a cosine similarity algorithm to obtain the second similarity, but is not limited thereto, and may also be based on other types of algorithms to calculate the second similarity, which is not limited in the embodiment of the present disclosure.

[0108] According to an embodiment of the present disclosure, the intention association weight and the second similarity may be processed based on a preset algorithm. For example, the second recognition result may be determined based on the product of the intention association weight and the second similarity.

[0109] In operation S250, target intention information is obtained according to the first recognition result and the second recognition result.

[0110] According to an embodiment of the present disclosure, a target recognition result can be obtained based on the first recognition result and the second recognition result, and when the target recognition result meets the requirements of a preset recognition condition, the initial intention information is determined as the target intention information.

[0111] According to an embodiment of the present disclosure, by determining a first similarity between target description text features and initial intent text features representing initial intent information, and determining a second similarity between a candidate intent corpus matching the description text and the description text, the similarity between the description text and the initial intent can be comprehensively considered from multiple dimensions of intent information and intent corpus, and then the target intent information can be identified based on the first recognition result and the second recognition result, thereby improving the recognition accuracy of the description text, reducing the probability of intent recognition errors, and achieving the technical effect of improving the accuracy of related resource recommendations.

[0112] According to an embodiment of the present disclosure, the intention information identification method also includes the following operations.

[0113] Obtain product display information and product operation information corresponding to the initial intent corpus, wherein the initial intent corpus is extracted from the user's historical query information; determine the initial intent information associated with the target product category based on the target product category represented by the product operation information; and determine the intent association weight between the initial intent corpus and the initial intent information based on the product display information and the product operation information.

[0114] According to an embodiment of the present disclosure, the user's historical query information may include files such as search logs and query logs in a historical time period that record the initial intent corpus provided by the user.

[0115] According to an embodiment of the present disclosure, the initial intent corpus may include text corpora such as product consultation texts and query texts provided by users in a historical time period. Product display information may include product recommendation resources displayed to users based on the initial intent corpus. Product operation information may be operation information for product display information corresponding to the initial intent corpus. Product operation information may include any type of operation such as browsing operation, collection operation, purchase operation, etc. for product display information. The embodiment of the present disclosure does not limit the specific operation type of product operation information, and those skilled in the art may select it according to actual needs. Accordingly, the product category of the product display information corresponding to the product operation information may be the target product category represented by the product operation information.

[0116] According to an embodiment of the present disclosure, the initial intention information may be associated with the target product category. For example, the initial intention information “recommendation of home appliance products” may be associated with the target product category “home appliance products”.

[0117] According to the embodiments of the present disclosure, a relationship mapping table can be established to construct the association relationship between the commodity category and the initial intention information, so that the corresponding initial intention information can be conveniently queried according to the target commodity category.

[0118] In one embodiment of the present disclosure, the intention association weight between the initial intention corpus and the initial intention information is determined, for example, it can be: when the input query information (initial intention corpus) "I want to buy laundry stuff" is detected, the commodity recommendation resources (i.e., commodity display information) are pushed to the client interface of the relevant user, and the commodity recommendation resources may include 100. By counting the commodity operation information for the commodity display information in the historical time period, for example, the frequency of click operation information for the "household daily necessities" commodity display information is counted as 80 times, it can be determined that the intention association weight between the initial corpus "I want to buy laundry stuff" and the initial intention information "household daily necessities commodity recommendation" can be 0.8. Correspondingly, when the frequency of click operation information for the "home appliances" commodity display information is counted as 20 times, it can be determined that the intention association weight between the initial intention corpus "I want to buy laundry stuff" and the initial intention information "home appliances commodity recommendation" can be 0.2.

[0119] It should be noted that, when there are multiple initial intent corpora corresponding to the same initial intent information, the intent association weights of the multiple initial intent corpora for the initial intent information can be normalized to facilitate the subsequent determination of the target intent information.

[0120] According to an embodiment of the present disclosure, by extracting the initial intent corpus from the user's historical query information in real time, updating the initial corpus in real time, and updating the intent association weight between the initial corpus and the initial intent information in real time, the similarity degree between the initial intent corpus and the initial intent information can be corrected in a timely manner, and it is facilitating the subsequent accurate determination of the second similarity, thereby improving the detection accuracy for the target intent information and reducing the probability of error in intent information recognition.

[0121] Figure 3 The flowchart of extracting text features of description text according to an embodiment of the present disclosure is schematically shown.

[0122] like Figure 3 As shown, in operation S210, extracting text features of the description text includes operations S310 to S340.

[0123] In operation S310, description keywords in the description text are extracted.

[0124] In operation S320, text features describing the keyword are extracted to obtain text features describing the keyword.

[0125] In operation S330, text features of the description text are extracted to obtain first description text features.

[0126] In operation S340, the first description text feature and the description keyword text feature are fused.

[0127] According to an embodiment of the present disclosure, description keywords in the description text can be extracted based on the word segmentation algorithm in the related art. The description keywords can include any type of characters or words in the description text. For example, each character in the description text can be used as a description keyword. The text features of the keywords can be extracted based on a neural network algorithm, or the description keywords can be encoded based on an encoding method such as a one-hot encoding algorithm to generate sparse text features of the description text. Accordingly, based on the same or similar method, the text features of the description text can be extracted to obtain the first description text features, that is, dense text features representing the description text can be generated.

[0128] According to an embodiment of the present disclosure, the fusion of the first description text feature and the description keyword text feature may be to fuse one or more description keyword text features as a whole with the first description text feature, or one or more description keyword text features may be fused individually with the first description text feature.

[0129] According to an embodiment of the present disclosure, the description keyword text features include multiple ones.

[0130] In operation S340, fusing the first description text feature and the description keyword text feature may include the following operations.

[0131] The plurality of description keyword text features are respectively fused with the first description text feature to obtain a second description text feature corresponding to the description keyword text feature; and based on the attention mechanism, the second description text features corresponding to each of the plurality of description keyword text features are fused.

[0132] According to an embodiment of the present disclosure, the description keyword text feature can be concatenated with the first description text feature to obtain a second description text feature corresponding to the description keyword text feature. The second description text feature is input into the attention network-based search to achieve the fusion of the second description text features corresponding to the multiple description keyword text features.

[0133] According to an embodiment of the present disclosure, an attention network can be constructed based on a Transformer network model, but is not limited thereto. An attention network can also be constructed based on other types of attention network algorithms, for example, an attention network can also be constructed based on a recurrent neural network algorithm.

[0134] According to an embodiment of the present disclosure, by fusing the second description text features corresponding to each of multiple description keyword text features based on an attention mechanism, the local semantic information represented by the description keyword text features can be fully integrated with the overall semantic information of the description text represented by the first description text features, so as to enhance the semantic richness of subsequent target description text features and avoid the loss of semantic information in the description text.

[0135] According to an embodiment of the present disclosure, in operation S230, determining the candidate intention corpus matching the description text may include the following operations.

[0136] Extracting intent keywords from the description text; and determining candidate intent corpora matching the intent keywords from an initial intent corpus set including the initial intent corpora.

[0137] According to an embodiment of the present disclosure, the intent keyword may include a word or phrase that can at least partially represent the intent semantic information in the description text. For example, if the description text is "I want to buy a washing machine", the intent keyword may include: "washing machine". The intent keyword may be matched with the initial intent corpus in the initial intent corpus set, and the initial corpus matching the intent keyword may be determined as a candidate intent corpus.

[0138] According to an embodiment of the present disclosure, the first recognition result includes a first recognition value, and the second recognition result includes a second recognition value.

[0139] In operation 250, obtaining target intention information according to the first recognition result and the second recognition result includes the following operations.

[0140] Determine a first target identification value based on the first identification value and the first identification weight; determine a second target identification value based on the second identification value and the second identification weight; determine a third target identification value based on the first target identification value and the second target identification value; and when the third target identification value is greater than a preset identification threshold, identify the initial intention information as target intention information.

[0141] According to the embodiments of the present disclosure, the first recognition result can be determined by a deep learning model, and accordingly, the first recognition weight can be determined according to the test results of the deep learning model, such as recall rate, accuracy rate, etc. Therefore, the numerical value relationship between the first target recognition value and the second target recognition value can be flexibly allocated according to the first recognition weight and the second recognition weight, so as to realize dynamic adjustment of the recognition of target intention information.

[0142] Figure 4 The application scenario diagram of the intent information recognition method according to an embodiment of the present disclosure is schematically shown.

[0143] like Figure 4As shown, the application scenario may include an intent recognition model 410, a first similarity detection model 420, and a second similarity detection model 430. The intent recognition model 410 may include a first text feature extraction layer 411 and a second text feature extraction layer 412. The first text feature extraction layer 411 may include a text encoding sublayer 4111, a forward feedback sublayer 4112, a target text feature extraction sublayer 4113, and a fully connected sublayer 4114.

[0144] The description text 401 is input into the text encoding sublayer 411, and the keyword text feature and the first description text feature can be output. The keyword text feature and the first description text feature can be input into the forward feedback sublayer 4112, and the second description text feature is output, thereby achieving a preliminary fusion between the keyword text feature and the first description text feature. The second description text feature is input into the target text feature extraction sublayer 4113 constructed based on the Transformer model, which can fully integrate the global semantic information and local semantic information in the description text. The fused features output by the target text feature extraction sublayer 4113 can be input into the fully connected sublayer 4114 to match the vector dimension, and the target description text feature is output.

[0145] The initial intent information 402 may be input to the second text feature extraction layer 412 of the intent recognition model 410 to output the initial intent text features. The second text feature extraction layer 412 may be constructed based on a multi-layer perceptron model.

[0146] The initial intention text features and the target description text features are input into the first similarity detection model 420 to obtain a first recognition result 441 (first recognition value). The first similarity detection model 420 can be constructed based on a cosine similarity algorithm.

[0147] In this embodiment, the description text 401 and the candidate intent corpus 403 may also be input into the second similarity detection model 430 to output a second recognition result 442. The candidate intent corpus 403 may be screened from the initial intent corpus set based on the intent keywords in the description text 401.

[0148] According to an embodiment of the present disclosure, the candidate intent corpus that matches the description text "I want to buy a washing machine" may be "Are there any discounts on washing machines?" or "Recommend a washing machine." The second similarity between the description text and "Are there any discounts on washing machines?" calculated by the Jaccard similarity algorithm is 0.33, and the second similarity between the description text and "Recommend a washing machine" is 0.6. Correspondingly, the intention association weights of the candidate intent corpus "Are there any discounts on washing machines?" and "Recommend a washing machine" with the initial intent information "Recommendations for household electrical appliance product categories" may be 0.8 and 0.6, respectively. Thus, the second recognition result (i.e., the second recognition value) of the description text "I want to buy a washing machine" for the initial intent information may be calculated to be 0.12 according to the calculation formula 0.8*0.33+0.6*0.625=0.64.

[0149] For example, the third target recognition value 451 may be determined based on the first recognition result 441 and the second recognition result 442. For example, the first target recognition value may be determined based on the product of the first recognition result 441 and the first recognition weight, the second target recognition value may be determined based on the product of the second recognition result 442 and the second recognition weight, and the third target recognition value 451 may be determined by giving the sum of the first target recognition value and the second target recognition value. When the third target recognition value 451 is greater than or equal to the preset recognition threshold, the initial intent information 402 is identified as the target intent information corresponding to the description text 410.

[0150] According to an embodiment of the present disclosure, the second recognition result 442 output by the second similarity detection model 430 can be used to correct the model prediction result of the first recognition result 441 obtained based on the intention recognition model 410, so as to avoid the problem of low prediction accuracy of the intention recognition model 410 due to imbalanced training data and poor training data quality, thereby improving the prediction accuracy of the target intent information.

[0151] According to an embodiment of the present disclosure, the intention information identification method also includes the following operations.

[0152] Push recommended resource information corresponding to the target intent information to the target client corresponding to the description text.

[0153] According to an embodiment of the present disclosure, when target intent information is identified, recommended resources matching the target intent information, such as product recommendation links, product marketing advertising information and other resources, can be pushed to the target client to meet the user's relevant needs in a timely manner.

[0154] Figure 5 The application scenario diagram of the intent information recognition method according to an embodiment of the present disclosure is schematically shown.

[0155] like Figure 5As shown, the training method of the deep learning model may include operations S510 to S540.

[0156] In operation S510, a training sample is obtained, wherein the training sample includes a sample intent corpus and a sample label, and the sample label includes sample intent information and a sample intent association weight corresponding to the sample intent information.

[0157] In operation S520, the sample intent corpus is input into the first text feature extraction layer of the initial deep learning model, and the predicted intent corpus text features are output.

[0158] In operation S530 , the sample intent information is input into the second text feature extraction layer of the initial deep learning model, and the sample intent text features are output.

[0159] In operation S540, an initial deep learning model is trained using the predicted intent corpus text features, the sample intent text features, and the sample intent association weights to obtain a trained deep learning model.

[0160] According to an embodiment of the present disclosure, the first text feature extraction layer can be constructed based on an attention network, but is not limited thereto. The first text feature extraction layer can also be constructed based on other types of attention network algorithms. For example, the first text feature extraction layer can also be constructed based on a recurrent neural network algorithm.

[0161] According to the embodiments of the present disclosure, the technical terms involved in the training method of the deep learning model, including but not limited to sample description text and sample intent information, may be the same or corresponding technical terms as those in the above-mentioned embodiments, such as description text, initial intent information, etc., and have the same or corresponding technical attributes, and the embodiments of the present disclosure will not be repeated here.

[0162] According to an embodiment of the present disclosure, in operation S540, training the initial deep learning model using the predicted intent corpus text features, sample intent text features, and sample intent association weights includes the following operations.

[0163] The initial loss function is updated according to the sample intent association weight to obtain the target loss function; the predicted intent corpus text features and the sample intent text features are input into the target loss function to obtain the target loss value; the parameters of the initial deep learning model are adjusted based on the target loss value to obtain the trained deep learning model.

[0164] According to an embodiment of the present disclosure, the product of the sample intention association weight and the initial loss function can be used as the target loss function. For example, the target loss function can be expressed by formula (1).

[0165] L1'(x1, x2)=A×L1(x1, x2); (1)

[0166] In formula (1), L1'() represents the target loss function, A represents the sample intent association weight, L1() represents the initial loss function, x1 can represent the predicted intent corpus text feature, and x2 represents the sample intent text feature.

[0167] According to the embodiments of the present disclosure, by updating the target loss function through the sample intent association weights, the deep learning model can fully learn the similarity between the sample intent corpus and the sample intent information, thereby improving the deep learning model's prediction degree of the similarity between the subsequent description text and the intent information, improving the prediction accuracy of the first recognition result representation similarity, and thereby improving the recognition accuracy of the target intent information.

[0168] According to an embodiment of the present disclosure, the above operation of updating the initial loss function according to the sample intention association weight includes the following operations.

[0169] The sample intention association weight is updated based on the preset rules to obtain the sample target weight; and the initial loss function is updated using the sample target weight.

[0170] According to an embodiment of the present disclosure, for example, when the sample intent association weight is a probability value between 0 and 1, the sample target weight can be obtained by summing the sample intent association weight and the preset probability modification positive value. For example, the preset probability modification positive value can be determined as 0.5, which means that the sample target weight representation value is between 0.5 and 1.5. In this way, the target loss function obtained by updating the sample target weight can enable the deep learning model to fully learn the similarity between the sample intent corpus with high similarity and the sample intent information through the output sample target loss value, thereby improving the prediction accuracy and prediction effect of the deep learning model.

[0171] Figure 6 A block diagram of an intention information recognition device according to an embodiment of the present disclosure is schematically shown.

[0172] like Figure 6 As shown, the intent information identification device 600 includes a text feature extraction module 610, a first recognition result determination module 620, a candidate intent corpus determination module 630, a second recognition result determination module 640 and a target intent information determination module 650.

[0173] The text feature extraction module 610 is used to extract text features of the description text to obtain target description text features;

[0174] The first recognition result determination module 620 is used to determine a first recognition result for the initial intent information according to a first similarity between the target description text feature and the initial intent text feature, wherein the initial intent text feature represents the initial intent information.

[0175] The candidate intent corpus determination module 630 is used to determine candidate intent corpus that matches the description text, wherein the candidate intent corpus is associated with an intent association weight corresponding to the initial intent information.

[0176] The second recognition result determination module 640 is used to determine a second recognition result for the initial intent information according to the intent association weight and the second similarity between the description text and the candidate intent corpus.

[0177] The target intention information determination module 650 is used to obtain the target intention information according to the first recognition result and the second recognition result.

[0178] According to an embodiment of the present disclosure, the intention information identification device further includes: an acquisition module, an initial intention information determination module and an intention association weight determination module.

[0179] The acquisition module is used to obtain product display information and product operation information corresponding to the initial intent corpus, wherein the initial intent corpus is extracted from the user's historical query information.

[0180] The initial intention information determination module is used to determine the initial intention information associated with the target commodity category according to the target commodity category represented by the commodity operation information.

[0181] The intention association weight determination module is used to determine the intention association weight between the initial intention corpus and the initial intention information based on the product display information and the product operation information.

[0182] According to an embodiment of the present disclosure, the candidate intention corpus determination module includes: an intention keyword extraction unit and a candidate intention corpus determination unit.

[0183] The intention keyword extraction unit is used to extract the intention keywords in the description text.

[0184] The candidate intention corpus determining unit is used to determine the candidate intention corpus matching the intention keyword from the initial intention corpus set including the initial intention corpus.

[0185] According to an embodiment of the present disclosure, the text feature extraction module includes: a description keyword extraction unit, a first text feature extraction unit, a second text feature extraction unit and a fusion unit.

[0186] The description keyword extraction unit is used to extract description keywords from the description text.

[0187] The first text feature extraction unit is used to extract text features describing keywords to obtain text features describing keywords.

[0188] The second text feature extraction unit is used to extract text features of the description text to obtain first description text features.

[0189] The fusion unit is used to fuse the first description text feature and the description keyword text feature.

[0190] According to an embodiment of the present disclosure, the description keyword text features include multiple ones.

[0191] The fusion unit includes: a first fusion subunit and a second fusion subunit.

[0192] The first fusion subunit is used to fuse the plurality of description keyword text features with the first description text feature respectively to obtain a second description text feature corresponding to the description keyword text feature.

[0193] The second fusion subunit is used to fuse the second description text features corresponding to each of the multiple description keyword text features based on the attention mechanism.

[0194] According to an embodiment of the present disclosure, the first recognition result includes a first recognition value, and the second recognition result includes a second recognition value.

[0195] Among them, the target intention information determination module includes: a first target identification value determination unit, a second target identification value determination unit, a third target identification value determination unit and a target intention information determination unit.

[0196] The first target recognition value determining unit is used to determine the first target recognition value according to the first recognition value and the first recognition weight.

[0197] The second target recognition value determination unit is used to determine the second target recognition value according to the second recognition value and the second recognition weight.

[0198] The third target identification value determining unit is used for determining a third target identification value according to the first target identification value and the second target identification value.

[0199] The target intention information determination unit is used to identify the initial intention information as target intention information when the third target recognition value is greater than a preset recognition threshold.

[0200] According to an embodiment of the present disclosure, the apparatus for identifying intention information further includes a push module.

[0201] The push module is used to push recommended resource information corresponding to the target intent information to the target client corresponding to the description text.

[0202] Figure 7 A block diagram of a deep learning model training device according to an embodiment of the present disclosure is schematically shown.

[0203] like Figure 7As shown, the deep learning model training device 700 includes: a training sample acquisition module 710, a prediction intent corpus text feature determination module 720, a sample intent text feature determination module 730 and a training module 740.

[0204] The training sample acquisition module 710 is used to acquire training samples, wherein the training samples include sample intent corpus and sample labels, and the sample labels include sample intent information and sample intent association weights corresponding to the sample intent information.

[0205] The predicted intent corpus text feature determination module 720 is used to input the sample intent corpus into the first text feature extraction layer of the initial deep learning model, and output the predicted intent corpus text features.

[0206] The sample intent text feature determination module 730 is used to input the sample intent information into the second text feature extraction layer of the initial deep learning model and output the sample intent text feature.

[0207] The training module 740 is used to train the initial deep learning model using the predicted intent corpus text features, sample intent text features and sample intent association weights to obtain a trained deep learning model.

[0208] According to an embodiment of the present disclosure, the training module includes: a target loss function obtaining unit, a target loss value obtaining unit and a training unit.

[0209] The target loss function obtaining unit is used to update the initial loss function according to the sample intention association weight to obtain the target loss function.

[0210] The target loss value obtaining unit is used to input the predicted intent corpus text features and the sample intent text features into the target loss function to obtain the target loss value.

[0211] The training unit is used to adjust the parameters of the initial deep learning model based on the target loss value to obtain a trained deep learning model.

[0212] According to an embodiment of the present disclosure, the target loss function obtaining unit includes: a sample target weight obtaining subunit and an updating subunit.

[0213] The sample target weight acquisition subunit is used to update the sample intention association weight based on preset rules to obtain the sample target weight.

[0214] The update subunit is used to update the initial loss function using the sample target weights.

[0215] According to the embodiments of the present invention, any one or more of the modules, units, and sub-units, or at least part of the functions of any one of them can be implemented in one module. According to the embodiments of the present invention, any one or more of the modules, units, and sub-units can be split into multiple modules for implementation. According to the embodiments of the present invention, any one or more of the modules, units, and sub-units can be at least partially implemented as hardware circuits, such as field programmable gate arrays (FPGAs), programmable logic arrays (PLAs), systems on chips, systems on substrates, systems on packages, application specific integrated circuits (ASICs), or can be implemented by hardware or firmware in any other reasonable way of integrating or packaging the circuit, or in any one of the three implementation methods of software, hardware, and firmware, or in any appropriate combination of any of them. Alternatively, according to the embodiments of the present invention, one or more of the modules, units, and sub-units can be at least partially implemented as computer program modules, and when the computer program modules are run, the corresponding functions can be performed.

[0216] For example, the text feature extraction module 610, the first recognition result determination module 620, the candidate intent corpus determination module 630, the second recognition result determination module 640 and the target intent information determination module 650, or any multiple of the training sample acquisition module 710, the predicted intent corpus text feature determination module 720, the sample intent text feature determination module 730 and the training module 740 can be combined in one module / unit / sub-unit for implementation, or any one of the modules / units / sub-units can be split into multiple modules / units / sub-units. Alternatively, at least part of the functions of one or more of these modules / units / sub-units can be combined with at least part of the functions of other modules / units / sub-units and implemented in one module / unit / sub-unit. According to an embodiment of the present disclosure, the text feature extraction module 610, the first recognition result determination module 620, the candidate intent corpus determination module 630, the second recognition result determination module 640 and the target intent information determination module 650, or at least one of the training sample acquisition module 710, the predicted intent corpus text feature determination module 720, the sample intent text feature determination module 730 and the training module 740 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application specific integrated circuit (ASIC), or can be implemented by hardware or firmware such as any other reasonable way of integrating or packaging the circuit, or can be implemented in any one of the three implementation methods of software, hardware and firmware or in an appropriate combination of any of them. Alternatively, the text feature extraction module 610, the first recognition result determination module 620, the candidate intent corpus determination module 630, the second recognition result determination module 640 and the target intent information determination module 650, or at least one of the training sample acquisition module 710, the predicted intent corpus text feature determination module 720, the sample intent text feature determination module 730 and the training module 740 can be at least partially implemented as a computer program module, which can perform corresponding functions when the computer program module is executed.

[0217] It should be noted that the intention information identification device part in the embodiment of the present disclosure corresponds to the intention information identification method part in the embodiment of the present disclosure. The description of the intention information identification device part specifically refers to the intention information identification method part, which will not be repeated here.

[0218] Figure 8 A block diagram of an electronic device suitable for implementing a method for identifying intent information and a method for training a deep learning model according to an embodiment of the present disclosure is schematically shown. Figure 8 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0219] like Figure 8 As shown, the electronic device 800 according to an embodiment of the present disclosure includes a processor 801, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 802 or a program loaded from a storage part 808 into a random access memory (RAM) 803. The processor 801 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or a related chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 801 may also include an onboard memory for caching purposes. The processor 801 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.

[0220] In RAM 803, various programs and data required for the operation of electronic device 800 are stored. Processor 801, ROM 802 and RAM 803 are connected to each other via bus 804. Processor 801 performs various operations of the method flow according to the embodiment of the present disclosure by executing the program in ROM 802 and / or RAM 803. It should be noted that the program can also be stored in one or more memories other than ROM 802 and RAM 803. Processor 801 can also perform various operations of the method flow according to the embodiment of the present disclosure by executing the program stored in the one or more memories.

[0221] According to an embodiment of the present disclosure, the electronic device 800 may further include an input / output (I / O) interface 805, which is also connected to the bus 804. The system 800 may further include one or more of the following components connected to the input / output (I / O) interface 805: an input portion 806 including a keyboard, a mouse, etc.; an output portion 807 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage portion 808 including a hard disk, etc.; and a communication portion 809 including a network interface card such as a LAN card, a modem, etc. The communication portion 809 performs communication processing via a network such as the Internet. A drive 810 is also connected to the input / output (I / O) interface 805 as needed. A removable medium 811, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 810 as needed, so that a computer program read therefrom is installed into the storage portion 808 as needed.

[0222] According to an embodiment of the present disclosure, the method flow according to an embodiment of the present disclosure can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a computer-readable storage medium, and the computer program contains a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 809, and / or installed from the removable medium 811. When the computer program is executed by the processor 801, the above-mentioned functions defined in the system of the embodiment of the present disclosure are executed. According to an embodiment of the present disclosure, the system, equipment, device, module, unit, etc. described above can be implemented by a computer program module.

[0223] The present disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or may exist independently without being assembled into the device / apparatus / system. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed, the method according to the embodiment of the present disclosure is implemented.

[0224] According to an embodiment of the present disclosure, the computer-readable storage medium may be a non-volatile computer-readable storage medium. For example, it may include, but is not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in combination with an instruction execution system, apparatus, or device.

[0225] For example, according to an embodiment of the present disclosure, the computer-readable storage medium may include the ROM 802 and / or the RAM 803 described above and / or one or more memories other than the ROM 802 and the RAM 803 .

[0226] An embodiment of the present disclosure also includes a computer program product, which includes a computer program, which contains program code for executing the method provided by the embodiment of the present disclosure. When the computer program product runs on an electronic device, the program code is used to enable the electronic device to implement the intent information recognition method and deep learning model training method provided by the embodiment of the present disclosure.

[0227] When the computer program is executed by the processor 801, the above functions defined in the system / device of the embodiment of the present disclosure are executed. According to the embodiment of the present disclosure, the system, device, module, unit, etc. described above can be implemented by a computer program module.

[0228] In one embodiment, the computer program may rely on tangible storage media such as optical storage devices, magnetic storage devices, etc. In another embodiment, the computer program may also be transmitted and distributed in the form of signals on a network medium, and downloaded and installed through the communication part 809, and / or installed from a removable medium 811. The program code contained in the computer program may be transmitted using any appropriate network medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

[0229] According to an embodiment of the present disclosure, the program code for executing the computer program provided by the embodiment of the present disclosure can be written in any combination of one or more programming languages. Specifically, these computing programs can be implemented using high-level process and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, Java, C++, python, "C" language or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, partially on the remote computing device, or entirely on the remote computing device or server. In the case of a remote computing device, the remote computing device can be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (for example, using an Internet service provider to connect through the Internet).

[0230] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram may represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of the code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box may also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of boxes in the block diagram or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions. It can be understood by those skilled in the art that the features recorded in the various embodiments and / or claims of the present disclosure can be combined and / or combined in a variety of ways, even if such a combination or combination is not explicitly recorded in the present disclosure. In particular, without departing from the spirit and teaching of the present disclosure, the features described in the various embodiments and / or claims of the present disclosure may be combined and / or combined in a variety of ways. All of these combinations and / or combinations fall within the scope of the present disclosure.

[0231] The embodiments of the present disclosure are described above. However, these embodiments are only for illustrative purposes and are not intended to limit the scope of the present disclosure. Although the embodiments are described above separately, this does not mean that the measures in the various embodiments cannot be used in combination to advantage. The scope of the present disclosure is defined by the attached claims and their equivalents. Without departing from the scope of the present disclosure, those skilled in the art may make a variety of substitutions and modifications, which should all fall within the scope of the present disclosure.

Claims

1. A method for identifying intent information, comprising: Extract the text features of the description text to obtain the target description text features; Determining a first recognition result for the initial intent information according to a first similarity between the target description text feature and the initial intent text feature, wherein the initial intent text feature represents the initial intent information; Determining a candidate intent corpus matching the description text, wherein the candidate intent corpus is associated with an intent association weight corresponding to the initial intent information; Determining a second recognition result for the initial intent information according to the intent association weight and a second similarity between the description text and the candidate intent corpus; Target intention information is obtained according to the first recognition result and the second recognition result.

2. The method according to claim 1, further comprising: Acquire product display information and product operation information corresponding to the initial intent corpus, wherein the initial intent corpus is extracted from the user's historical query information; Determining initial intention information associated with the target commodity category according to the target commodity category represented by the commodity operation information; and An intention association weight between the initial intention corpus and the initial intention information is determined according to the product display information and the product operation information.

3. The method according to claim 2, wherein: The determining of the candidate intent corpus matching the description text comprises: Extracting intent keywords from the description text; and Determine a candidate intent corpus matching the intent keyword from an initial intent corpus set including the initial intent corpus.

4. The method according to claim 1, wherein: The text features of the extracted description text include: Extracting description keywords from the description text; Extracting text features of the description keyword to obtain text features of the description keyword; Extracting text features of the description text to obtain first description text features; and The first description text feature and the description keyword text feature are fused.

5. The method according to claim 4, wherein: The description keyword text features include multiple; The step of fusing the first description text feature and the description keyword text feature comprises: Merging the plurality of description keyword text features with the first description text feature respectively to obtain a second description text feature corresponding to the description keyword text feature; and Based on the attention mechanism, the second description text features corresponding to each of the plurality of description keyword text features are fused.

6. The method according to claim 1, wherein: The first recognition result includes a first recognition value, and the second recognition result includes a second recognition value; Wherein, obtaining target intention information according to the first recognition result and the second recognition result includes: Determining a first target recognition value according to the first recognition value and the first recognition weight; Determining a second target recognition value according to the second recognition value and a second recognition weight; determining a third target identification value based on the first target identification value and the second target identification value; and When the third target recognition value is greater than a preset recognition threshold, the initial intention information is recognized as target intention information.

7. The method according to any one of claims 1 to 6, further comprising: The recommended resource information corresponding to the target intent information is pushed to the target client corresponding to the description text.

8. A method for training a deep learning model, comprising: Acquire a training sample, wherein the training sample includes a sample intent corpus and a sample label, and the sample label includes sample intent information and a sample intent association weight corresponding to the sample intent information; Inputting the sample intent corpus into the first text feature extraction layer of the initial deep learning model, and outputting the predicted intent corpus text features; Inputting the sample intent information into the second text feature extraction layer of the initial deep learning model, and outputting the sample intent text features; The initial deep learning model is trained using the predicted intent corpus text features, the sample intent text features and the sample intent association weights to obtain a trained deep learning model.

9. The method according to claim 8, wherein: The training of the initial deep learning model using the predicted intent corpus text features, the sample intent text features and the sample intent association weights comprises: Update the initial loss function according to the sample intention association weight to obtain the target loss function; Inputting the predicted intent corpus text features and the sample intent text features into the target loss function to obtain a target loss value; The parameters of the initial deep learning model are adjusted based on the target loss value to obtain the trained deep learning model.

10. The method according to claim 9, wherein: The updating of the initial loss function according to the sample intention association weight comprises: Update the sample intention association weight based on preset rules to obtain a sample target weight; and The initial loss function is updated using the sample target weights.

11. An intention information recognition device, comprising: A text feature extraction module is used to extract text features of the description text to obtain target description text features; A first recognition result determination module, configured to determine a first recognition result for the initial intent information according to a first similarity between the target description text feature and the initial intent text feature, wherein the initial intent text feature represents the initial intent information; A candidate intent corpus determination module, used to determine a candidate intent corpus matching the description text, wherein the candidate intent corpus is associated with an intent association weight corresponding to the initial intent information; A second recognition result determination module, configured to determine a second recognition result for the initial intent information according to the intent association weight and a second similarity between the description text and the candidate intent corpus; A target intention information determination module is used to obtain target intention information based on the first recognition result and the second recognition result.

12. A training device for a deep learning model, comprising: A training sample acquisition module, used to acquire training samples, wherein the training samples include sample intent corpus and sample labels, and the sample labels include sample intent information and sample intent association weights corresponding to the sample intent information; A predicted intent corpus text feature determination module, used to input the sample intent corpus into the first text feature extraction layer of the initial deep learning model, and output the predicted intent corpus text features; A sample intention text feature determination module, used for inputting the sample intention information into the second text feature extraction layer of the initial deep learning model, and outputting the sample intention text feature; A training module is used to train the initial deep learning model using the predicted intent corpus text features, the sample intent text features and the sample intent association weights to obtain a trained deep learning model.

13. An electronic device, comprising: one or more processors; a memory for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 10.

14. A computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, enables the processor to implement the method according to any one of claims 1 to 10.

15. A computer program product comprising a computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 10.

Images