Method, device and electronic device for identifying user intention based on feedback content

Abstract

The embodiments of the present invention provide a method, device and electronic device for identifying user intent for feedback content, which are applied to the field of natural language processing technology. The method includes: obtaining the target feedback content to be identified; based on the preset correspondence between each classification condition and the intent category, determining the intent category corresponding to the classification condition satisfied by the target feedback content as the initial category; wherein the intent category corresponding to each classification condition is the intent category that can be represented by the feedback content that satisfies the classification condition; if the initial category is judged to be a specified category, the intent recognition result of the target feedback content is determined using a pre-trained intent classification model; otherwise, the initial category is determined as the intent recognition result of the target feedback content. Through this solution, the efficiency of determining the user's feedback intent based on the user's feedback content can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of natural language processing, and in particular to a method, device and electronic device for identifying user intention for feedback content. Background Art

[0002] In order to better improve user experience, product suppliers often set up product feedback centers to receive user feedback on their products. Staff need to determine the user's feedback intention based on the user's feedback content, such as whether the user is dissatisfied with a certain function of the product or the user is seeking help, and then they can handle it in a targeted manner based on the user's feedback intention.

[0003] In the prior art, staff are required to determine the user's feedback intention based on the user's feedback content, which is inefficient. Summary of the invention

[0004] The purpose of the embodiment of the present invention is to provide a method for identifying user intentions for feedback content, so as to improve the efficiency of determining the user's feedback intentions based on the user's feedback content. The specific technical solution is as follows:

[0005] In a first aspect, an embodiment of the present invention provides a method for identifying user intent for feedback content, comprising:

[0006] Obtain target feedback content to be identified;

[0007] Based on the preset correspondence between each classification condition and the intention category, determine the intention category corresponding to the classification condition satisfied by the target feedback content as the initial category; wherein the intention category corresponding to each classification condition is the intention category that can be represented by the feedback content satisfying the classification condition;

[0008] If it is determined that the initial category is a specified category, using a pre-trained intent classification model to determine an intent recognition result of the target feedback content;

[0009] Otherwise, determining the initial category as the intention recognition result of the target feedback content;

[0010] Among them, the designated category is an intent category for which a similar intent category exists; the intent classification model is a classification model trained based on multiple sample feedback contents, and the multiple sample feedback contents include: sample feedback contents that can characterize the designated category, and sample feedback contents that can characterize similar intent categories of the designated category.

[0011] Optionally, before determining, based on a preset correspondence between each classification condition and the intention category, the intention category corresponding to the classification condition satisfied by the target feedback content as the initial category, the method further includes:

[0012] Generate a sentence vector representing the target feedback content as a target vector;

[0013] Calculating the distance between the target vector and each cluster in the cluster space, wherein the cluster space is established based on the sentence vector of the sample feedback content, and each cluster in the cluster space is associated with an intent category;

[0014] Determine a target cluster whose distance to the target vector is less than a preset threshold, and determine an intent category associated with the target cluster as a pre-selected category;

[0015] The determining, based on the preset correspondence between each classification condition and the intention category, the intention category corresponding to the classification condition satisfied by the target feedback content as the initial category includes:

[0016] Based on the preset correspondence between each classification condition and the intention category and the pre-selected category, the intention category corresponding to the classification condition satisfied by the target feedback content is determined as the initial category.

[0017] Optionally, the determining, based on the preset correspondence between each classification condition and the intention category and the preselected category, the intention category corresponding to the classification condition satisfied by the target feedback content as the initial category includes:

[0018] From the preset correspondence between each classification condition and the intended category, searching for the classification condition corresponding to the pre-selected category;

[0019] The classification conditions satisfied by the target feedback content are determined from the searched classification conditions, and based on the corresponding relationship, the intention category corresponding to the classification conditions satisfied by the target feedback content is determined as the initial category.

[0020] Optionally, generating a sentence vector representing the target feedback content includes:

[0021] Determining key participles contained in the target feedback content, wherein the key participles are participles belonging to a preset participle type;

[0022] Generate a word vector for the key word segmentation, and generate a sentence vector representing the target feedback content based on the word vector for the key word segmentation.

[0023] Optionally, the intent classification model is: an intent classification model trained based on a random forest classification model.

[0024] Optionally, the method further includes:

[0025] According to a preset mapping relationship between a preset segmentation word and a subcategory under the intention recognition result, a subcategory having a mapping relationship with the preset segmentation word included in the target feedback content is determined as a subcategory of the target feedback content.

[0026] In a second aspect, an embodiment of the present invention provides a device for identifying user intention for feedback content, comprising:

[0027] A content acquisition module, used to acquire target feedback content to be identified;

[0028] A category determination module, for determining, based on a preset correspondence between each classification condition and the intention category, the intention category corresponding to the classification condition satisfied by the target feedback content as the initial category; wherein the intention category corresponding to each classification condition is the intention category that can be represented by the feedback content satisfying the classification condition;

[0029] A result determination module, for determining the intention recognition result of the target feedback content by using a pre-trained intention classification model if the initial category is determined to be a specified category; otherwise, determining the initial category as the intention recognition result of the target feedback content;

[0030] Among them, the designated category is an intent category for which a similar intent category exists; the intent classification model is a classification model trained based on multiple sample feedback contents, and the multiple sample feedback contents include: sample feedback contents that can characterize the designated category, and sample feedback contents that can characterize similar intent categories of the designated category.

[0031] Optionally, the device further comprises:

[0032] a vector generating module, configured to generate a sentence vector representing the target feedback content as a target vector before the category determining module determines the intent category corresponding to the classification condition satisfied by the target feedback content as an initial category based on the preset correspondence between each classification condition and the intent category;

[0033] A distance calculation module, used to calculate the distance between the target vector and each cluster in the cluster space, wherein the cluster space is established based on the sentence vector of the sample feedback content, and each cluster in the cluster space is associated with an intent category;

[0034] A cluster determination module, used to determine a target cluster whose distance to the target vector is less than a preset threshold, and determine an intent category associated with the target cluster as a pre-selected category;

[0035] The category determination module is specifically used to determine the intention category corresponding to the classification condition satisfied by the target feedback content as the initial category based on the preset correspondence between each classification condition and the intention category and the pre-selected category.

[0036] Optionally, the category determination module is specifically used to search for classification conditions corresponding to the preselected category from preset correspondences between various classification conditions and intention categories, and determine the classification conditions satisfied by the target feedback content from the searched classification conditions, and based on the correspondence, determine the intention category corresponding to the classification conditions satisfied by the target feedback content as the initial category.

[0037] Optionally, the vector generation module is specifically used to determine the key participles contained in the target feedback content, wherein the key participles are participles belonging to a preset participle type, and generate word vectors for the key participles, and based on the word vectors for the key participles, generate sentence vectors representing the target feedback content.

[0038] Optionally, the intent classification model is: an intent classification model trained based on a random forest classification model.

[0039] Optionally, the device further comprises:

[0040] A subcategory determination module is used to determine a subcategory that has a mapping relationship with the preset segmentation contained in the target feedback content as a subcategory of the target feedback content according to a preset mapping relationship between the preset segmentation and the subcategory under the intention recognition result.

[0041] In a third aspect, an embodiment of the present invention provides an electronic device, including a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus;

[0042] Memory, used to store computer programs;

[0043] The processor is used to implement the method steps provided in the first aspect of claim 1 when executing the program stored in the memory.

[0044] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method steps provided in the first aspect are implemented.

[0045] In the method for identifying user intent for feedback content provided by the embodiment of the present invention, since the correspondence between each classification condition and the intent category is pre-established, the intent category corresponding to the classification condition satisfied by the target feedback content can be determined, and further, when the initial category is a specified category, the intent classification model is used to determine the intent identification result of the target feedback content, otherwise the initial category is determined as the intent identification result of the target feedback content. It can be seen that this solution can avoid manually determining the user's feedback intent based on the user's feedback content, and realizes automatic identification of the user's intent of the target feedback content, thereby improving the identification efficiency.

[0046] In addition, this solution adopts the correspondence between each classification condition and the intent category, and combines the intent classification model, so that when the initial category obtained by the correspondence relationship has similar categories, the initial category can be corrected, thereby further ensuring the accuracy of intent recognition. Of course, any product or method implementing the present invention does not necessarily need to achieve all the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art are briefly introduced below.

[0048] Figure 1 A flowchart of a method for identifying user intentions for feedback content provided by an embodiment of the present invention;

[0049] Figure 2 Another flow chart of a method for identifying user intent for feedback content provided by an embodiment of the present invention;

[0050] Figure 3 A schematic diagram of the structure of a device for identifying user intentions for feedback content provided by an embodiment of the present invention;

[0051] Figure 4 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0052] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0053] In order to improve the efficiency of determining a user's feedback intention based on the user's feedback content, embodiments of the present invention provide a method, device, and electronic device for identifying a user's intention with respect to feedback content.

[0054] The method for identifying user intention for feedback content provided by the embodiment of the present invention is applied to an electronic device. In a specific application, the electronic device may be a server, or a terminal device such as a smart phone or a tablet computer.

[0055] In one embodiment of the present invention, a method for identifying user intentions for feedback content is provided. Figure 1 As shown, the method comprises the following steps:

[0056] S101, obtaining target feedback content to be identified.

[0057] In this step, the electronic device can receive the user's feedback content on the product, that is, the feedback text, and use the received feedback content as the target feedback content. Of course, the electronic device can also obtain the target feedback content from the feedback center. Among them, the feedback center is a data center for collecting and storing user feedback content such as comments, suggestions and questions on the product. Therefore, the electronic device can read the feedback content from the database in the feedback center for storing user feedback content on the product, and use the read feedback content as the target feedback content.

[0058] The above target feedback content may be user comments, opinions, questions, etc. For example, for cleaning tool application software, the target feedback content may be "how to clean application garbage", "what to do if files are accidentally deleted", etc.

[0059] The target feedback content may also be feedback content obtained by preprocessing the received and acquired feedback content, wherein the preprocessing is used to remove special characters such as emoticons, symbols, and meaningless repeated words contained in the feedback content.

[0060] S102, based on the preset correspondence between each classification condition and the intention category, determine the intention category corresponding to the classification condition satisfied by the target feedback content as the initial category; wherein the intention category corresponding to each classification condition is the intention category that can be represented by the feedback content satisfying the classification condition.

[0061] Among them, the intent category can be pre-set based on the functional points of the product. For example, taking the chat software as an example, the opinion category can include chat category, address book category and group chat category, etc. Of course, the intent category can also be determined based on the collected sample feedback content.

[0062] In order to automatically identify the intent category, a large amount of feedback content can be collected in advance, and the intent categories represented by the large amount of feedback content and the text information of the feedback content itself can be used to establish a correspondence between each classification condition and the intent category. The intent category corresponding to each classification condition is the intent category that can be represented by the feedback content that meets the classification condition. Based on this correspondence, when it is necessary to identify the user intent of the target feedback content, the classification condition satisfied by the target feedback content can be determined in the pre-set classification conditions, and then the intent category corresponding to the classification condition satisfied by the target feedback content is determined as the initial category. The classification condition corresponding to each intent category can be set based on a keyword, wherein the keyword is related to the feedback content that can represent the intent category, such as the vocabulary that the feedback content that can represent the intent category may contain. At this time, the correspondence between each classification condition and the intent category can be called a keyword map, and the corresponding intent category can be determined through the classification conditions for keywords in the keyword map.

[0063] For example, the intent categories include interface display categories and file processing categories. When the words "icon", "font", and "background" appear in the feedback content, the feedback content is likely to be in the interface display category. Therefore, "icon", "font", and "background" are used as keywords to set the classification conditions corresponding to the interface display category. For example, the classification conditions corresponding to the interface display category can be: the feedback content contains "background", then when the target feedback content contains "background", it is determined that the target feedback content belongs to the interface display category.

[0064] In addition, the setting of the classification conditions corresponding to each intent category can also be set in combination with different dimensional information such as the sentence structure of the feedback content, the emotional tendency of the feedback content, the position of the keywords in the feedback content, actual needs, etc., and the present invention does not make specific limitations on this.

[0065] S103, if it is determined that the initial category is the specified category, use the pre-trained intent classification model to determine the intent recognition result of the target feedback content; otherwise, the initial category is determined as the intent recognition result of the target feedback content; wherein the specified category is an intent category for which a similar intent category exists; the intent classification model is a classification model trained based on multiple sample feedback contents, and the multiple sample feedback contents include: sample feedback contents that can represent the specified category, and sample feedback contents of similar intent categories that can represent the specified category.

[0066] Among them, since the classification conditions and corresponding relationships are set manually based on experience and needs, and the feedback opinions that humans have been exposed to are limited, the set classification conditions may not cover all possible situations. In particular, for multiple similar intent categories, it often leads to the target feedback content literally meeting a certain classification condition, but the intent category actually represented may be a similar category to the intent category corresponding to the classification condition.

[0067] In order to improve the accuracy of feedback intent recognition, a relatively similar intent category can be used as a designated category, and then a pre-trained intent classification model is used to determine the intent recognition result of the target feedback content to correct the initial category. If the initial category is determined to be a designated category, the pre-trained intent classification model is used to determine the intent recognition result of the target feedback content, specifically: the target feedback content is input into the pre-trained intent classification model to obtain the intent recognition result of the target feedback content. When the initial category is not a designated category, the initial category obtained by step S102 is also accurate, so the initial category can be determined as the intent recognition result of the target feedback content.

[0068] For example, if the desktop lock category and the application lock category are two similar intent categories, then both the desktop lock category and the application lock category can be used as designated categories. When the desktop lock category is used as the designated category, the similar category is the application category, and when the application lock category is used as the designated category, the similar category is the desktop category. When the initial category is the desktop lock category, the intent classification model is used to re-determine whether the target feedback content represents the desktop lock category or the application category.

[0069] Optionally, intent classification models for respective groups of categories may be preset, wherein each group of categories includes a plurality of similar intent categories.

[0070] In one implementation, the method of training the classification model based on multiple sample feedback contents may include:

[0071] Acquire multiple sample feedback contents of users for the product collected in advance, and the calibration contents of each feedback content; wherein each sample feedback content includes: sample feedback contents that can represent a specified category, and sample feedback contents of a similar intent category that can represent the specified category, and the calibration contents of each sample feedback content are the intent category that can be represented by the sample feedback content, and the calibration contents can be obtained by manual determination or other methods;

[0072] The initial intent classification model is trained using each sample feedback content and calibration content as training data. Specifically: each sample feedback content is input into the initial intent classification model to obtain the intent recognition result of the sample feedback content; based on the difference between the intent recognition result of each sample feedback content and the calibration content, the loss value is calculated; if the loss value is less than the predetermined loss threshold, the intent classification model is determined to have converged, and the trained intent classification model is obtained. If the loss value is not less than the predetermined loss value, the network parameters of the intent classification model are adjusted, and the step of inputting each sample feedback content into the initial intent classification model is returned to obtain the intent recognition result of the sample feedback content, so as to continue training the intent classification model. Among them, regarding the method of generating sample feedback content, you can refer to the method of generating target feedback content, which will not be elaborated here.

[0073] The above-mentioned intent classification model can be an intent classification model trained based on a random forest classification model. The random forest classification model is composed of a series of independent decision trees, and each decision tree constitutes the minimum component of the entire random forest algorithm. When the random forest classification model is given a data to be classified, each decision tree will make independent judgments on the input without affecting each other, and finally select the optimal classification result of the entire classifier by voting. The decision-making ability of a single decision tree is often weak, but when a series of decision trees are combined, their decision-making ability will be very strong.

[0074] In the solution provided by this embodiment, the correspondence between each classification condition and the intent category can be established in advance, and then the intent category corresponding to the classification condition satisfied by the target feedback content can be determined, and further when the initial category is a specified category, the intent classification model is used to determine the intent recognition result of the target feedback content, otherwise the initial category is determined as the intent recognition result of the target feedback content. It can be seen that this solution can avoid manually determining the user's feedback intent based on the user's feedback content, and realizes automatic recognition of the user's intent of the target feedback content, thereby improving the recognition efficiency.

[0075] In addition, this solution adopts the correspondence between each classification condition and the intent category, and combines it with the intent classification model. When the initial category obtained through the correspondence has similar categories, the initial category can be corrected, thereby further ensuring the accuracy of intent recognition.

[0076] based on Figure 1 Examples of Figure 2 As shown, the method for identifying user intention for feedback content provided by another embodiment of the present invention may further include, before S102:

[0077] S104, generating a sentence vector representing the target feedback content as a target vector;

[0078] The sentence vector representing the target feedback content can be generated based on the word vectors of the word segments contained in the target feedback content. Specifically, the word segments of the target feedback content can be obtained by text segmentation, etc., and then the word vector of each word segment can be determined, and then the sentence vector representing the target feedback content can be generated by the determined word vectors.

[0079] S105, calculating the distance between the target vector and each cluster in the cluster space, wherein the cluster space is established based on the sentence vector of the sample feedback content, and each cluster in the cluster space is associated with an intent category;

[0080] Among them, it can be known from the above embodiments that the intent category corresponding to each classification condition can be determined based on the collected sample feedback content. Specifically, the sentence vectors of the collected sample feedback content can be clustered to generate a clustering space, and multiple clusters after clustering can be obtained. Further, the associated intent category can be set for each cluster based on the actual usage scenario and needs.

[0081] After determining the target vector of the target feedback content, the distance between the target vector and each cluster can be calculated in the cluster space. The distance between the target vector and each cluster can be the cosine distance between the target adjacent vector and the centroid of each cluster. The centroid of each cluster is the mean of the sentence vectors contained in each cluster.

[0082] S106, determining a target cluster whose distance to the target vector is less than a preset threshold, and determining an intention category associated with the target cluster as a pre-selected category.

[0083] The preset threshold value may be determined based on demand and experience.

[0084] Exemplarily, the cluster space contains three clusters, namely cluster A, cluster B and cluster C, and the intent category associated with cluster A is intent category A, the intent category associated with cluster B is intent category B, and the intent category associated with cluster C is intent category C. Through calculation, it is determined that the distance between the target neighbor and cluster A is 0.2, the distance between the target neighbor and cluster B is 0.4, and the distance between the target neighbor and cluster C is 0.6. When the preset threshold is 0.5, cluster A and cluster B can be determined as target clusters. And further determine that intent category A and intent category are pre-selected categories.

[0085] Accordingly, after obtaining the pre-categorization, in one implementation, the above step S102 may include the following steps:

[0086] S102A, based on the preset correspondence between each classification condition and the intention category and the pre-selected category, determine the intention category corresponding to the classification condition satisfied by the target feedback content as the initial category.

[0087] Among them, optionally, in one implementation, in order to improve the accuracy of user intent recognition, the classification conditions satisfied by the target feedback content may be determined from various classification conditions, and when there are multiple classification conditions satisfied by the target feedback content, an initial category may be determined from the multiple classification conditions determined based on a pre-selected category. For example, the intent category corresponding to each of the multiple classification conditions may be determined first, and it may be determined whether there is a pre-selected category in the determined intent category, and if so, the pre-selected category may be used as the initial category.

[0088] Exemplarily, the preset classification conditions include condition 1, condition 2, and condition 3, the intent category corresponding to condition 1 is category 1, the intent category corresponding to condition 2 is category 2, and the intent category corresponding to condition 3 is category 3. When the target feedback content satisfies both condition 1 and condition 2, and the pre-selected category is category 1, category 1 can be used as the initial category.

[0089] Optionally, in another implementation, in order to improve the efficiency of user intent recognition, the initial category may be determined in the following manner:

[0090] From the preset correspondence between each classification condition and the intention category, search for the classification condition corresponding to the preselected category; from the searched classification conditions, determine the classification conditions satisfied by the target feedback content, and based on the correspondence, determine the intention category corresponding to the classification conditions satisfied by the target feedback content as the initial category.

[0091] Exemplarily, the preset classification conditions include condition 1, condition 2, and condition 3. The intent category corresponding to condition 1 is category 1, the intent category corresponding to condition 2 is category 2, and the intent category corresponding to condition 3 is category 3. When the pre-selected categories are category 1 and category 2, the classification conditions satisfied by the target feedback content can be determined from condition 1 and condition 2, and the intent category corresponding to the classification conditions satisfied by the target feedback content is used as the initial category.

[0092] In the solution provided in this embodiment, pre-selected categories that the target feedback content may satisfy can be screened out from various intent categories based on the sentence vector of the target feedback content, and the initial category of the target feedback content can be further determined in combination with the pre-selected categories and classification conditions, thereby improving the accuracy and efficiency of user intent recognition.

[0093] Optionally, in another embodiment of the present invention, the above step S104 may include the following steps:

[0094] Determine key participles included in the target feedback content, wherein the key participles are participles belonging to a preset participle type; generate word vectors for the key participles, and generate sentence vectors representing the target feedback content based on the word vectors for the key participles.

[0095] Before determining the key word segmentation, data processing may be performed on the target feedback content to remove special characters such as emoticons in the target feedback content.

[0096] The above-mentioned key words are words belonging to preset types, wherein the preset types can be determined according to actual needs and experience. For example, the preset types may include functional types, emotional types, and request types, wherein the functional type words may be words describing the functional points of the application, such as "interface", "application lock", "junk cleaning", etc. The emotional type words may be words describing emotional tendencies, such as "very good", "good", "like", "very bad", "too bad", "dislike", etc. The request type words may be words describing the required requests, such as "what to do", "how to set", "how to restore", etc.

[0097] In order to accurately determine the key word segmentation contained in the target feedback content, a key dictionary is pre-established, wherein the key dictionary contains preset types of segmentation. The key dictionary can be established based on the Jieba segmentation custom dictionary. For example, the custom keyword part in the Jieba segmentation custom dictionary is used as the key dictionary. After the target feedback content is obtained, the segmentation that appears in the keyword dictionary in the target feedback content can be found, which is the key word segmentation of the target feedback content.

[0098] The dimension of the word vector generated above can be determined based on experience and scenarios. After the dimension of the word vector is determined in advance, the word vector of the key word segmentation can be generated in a variety of ways. For example, the word vector of the key word segmentation can be generated in one of two ways:

[0099] The first method: You can use the TF-IDF (Term Frequency-Inverse Document Frequenc y) algorithm to generate word vectors for key word segments.

[0100] The second method is to generate word vectors for key word segments based on a word vector model. The word vector model can be a word2vec word vector model or a GloVe (Global vectors for word representation) word vector model. The choice of model can be determined based on actual needs.

[0101] After the word vector of the key word is generated, a sentence vector representing the target feedback content may be generated based on the word vector, wherein the sentence vector representing the target feedback content has the same dimension as the word vector of the key word.

[0102] Optionally, in one implementation, the sentence vector of the target feedback content may be calculated based on the word vectors of the key word and the word vectors of the non-key word, wherein the non-key word is the word included in the target feedback content except the key word.

[0103] Among them, a keyword dictionary that introduces the Jieba word segmentation custom dictionary can be pre-built, and the keyword dictionary contains custom key word segmentations, which can be determined in combination with the actual use scenario, experience and sample set. At the same time, there is also a general dictionary in the Jieba word segmentation custom dictionary. After obtaining the target feedback content, the key word segmentations contained in the Jieba word segmentation custom dictionary and the non-key word segmentations contained in the general dictionary can be screened out.

[0104] Then, word vectors of key word segments and word vectors of non-key word segments are generated according to the same dimension, and the word vectors of key word segments and word vectors of non-key word segments can be operated between vectors to calculate the sentence vector of the target feedback content.

[0105] Optionally, the sentence vector of the target feedback content can be calculated in one of the following two ways:

[0106] The first method is: calculate the product of the word length of the target feedback content and the word vector of the key word to obtain the first word vector; perform weighted average on the first word vector and the word vector of the non-key word to generate the sentence vector of the target feedback content.

[0107] In this method, the word length of the target feedback content is the number of segmented words of the target feedback content, that is, the sum of the number of key segmented words and non-key segmented words. By multiplying the word length with the word vector of the key segmented word, the proportion of the word vector of the key segmented word in the sentence vector can be increased, and the emotional tendency carried by the key segmented word can be prevented from being covered by the non-key segmented word, making the subsequent user intent recognition more accurate.

[0108] The second method: calculate the product of a preset multiple, the word length of the target feedback content, and the word vector of the key word segmentation to obtain a second word vector; perform weighted averaging on the second word vector and the word vector of the non-key word segmentation to generate a sentence vector of the target feedback content.

[0109] In this method, the proportion of the word vector of the key word segment in the sentence vector is further increased by the preset multiple, and the accuracy of user intent recognition is further improved. Among them, the preset multiple can be determined based on actual needs and experience, such as 3 times.

[0110] Optionally, in another embodiment of the present invention, the method for identifying user intentions for feedback content may further include:

[0111] According to a preset mapping relationship between a preset segmentation word and a subcategory under an intention recognition result, a subcategory having a mapping relationship with the preset segmentation word included in the target feedback content is determined as a subcategory of the target feedback content.

[0112] Among them, the intent category can contain multiple subcategories. For example, the intent category of cleaning can contain file cleaning subcategories, image cleaning subcategories, and video cleaning subcategories. Different subcategories are used to further segment the user intent of the feedback content, making the user intent identification more accurate.

[0113] Each subcategory that an intent category can contain has a mapping relationship established in advance with a preset segmentation word. Under the same intent category, the preset segmentations mapped to different subcategories are different from each other. The preset segmentations of each subcategory can be recorded by establishing a subcategory keyword dictionary.

[0114] Exemplarily, the intent category is taken as an example of a cleaning category. Among them, each subcategory that the cleaning category can contain has a mapping relationship established in advance with a preset segmentation word. Under the same intent category, the preset segmentations mapped to different subcategories are different from each other. For example, the intent category is a cleaning category, and the cleaning category includes three subcategories: a file cleaning subcategory, a picture cleaning subcategory, and a video cleaning subcategory. Among them, the preset segmentation word that has a mapping relationship with the file cleaning subcategory can be "file", "manuscript", etc., the preset segmentation word that has a mapping relationship with the picture cleaning subcategory can be "picture", "photo", etc., and the preset segmentation word that has a mapping relationship with the video cleaning subcategory can be "video", "movie", etc. When the target feedback content contains "video", the video cleaning subcategory can be determined as a subcategory of the target feedback content.

[0115] In the solution provided in this embodiment, by presetting the mapping relationship between the word segmentation and the subcategories under the intention recognition result, the subcategories of the target feedback content can be determined based on the intention recognition result. Yes, the target feedback content can be further subdivided, so that the user intention can be more accurately identified.

[0116] Corresponding to the user intention recognition method for feedback content provided in the above embodiment, Figure 3 As shown, an embodiment of the present invention further provides a device for identifying user intention for feedback content, the device comprising:

[0117] Content acquisition module 301, used to acquire target feedback content to be identified;

[0118] The category determination module 302 is used to determine the intent category corresponding to the classification condition satisfied by the target feedback content as the initial category based on the preset correspondence between each classification condition and the intent category; wherein the intent category corresponding to each classification condition is the intent category that can be represented by the feedback content satisfying the classification condition;

[0119] The result determination module 303 is used to determine the intention recognition result of the target feedback content by using the pre-trained intention classification model if the initial category is determined to be the specified category; otherwise, the initial category is determined to be the intention recognition result of the target feedback content;

[0120] Among them, the designated category is an intent category in which a similar intent category exists; the intent classification model is a classification model trained based on multiple sample feedback contents, and the multiple sample feedback contents include: sample feedback contents that can represent the designated category, and sample feedback contents that can represent similar intent categories of the designated category.

[0121] Furthermore, the device also includes:

[0122] A vector generation module, used for generating a sentence vector representing the target feedback content as a target vector before the category determination module executes the predetermined correspondence between each classification condition and the intention category to determine the intention category corresponding to the classification condition satisfied by the target feedback content as the initial category;

[0123] A distance calculation module is used to calculate the distance between the target vector and each cluster in the cluster space, wherein the cluster space is established based on the sentence vector of the sample feedback content, and each cluster in the cluster space is associated with an intent category;

[0124] A cluster determination module, used to determine a target cluster whose distance to the target vector is less than a preset threshold, and determine the intent category associated with the target cluster as a pre-selected category;

[0125] The category determination module is specifically used to determine the intent category corresponding to the classification condition satisfied by the target feedback content as the initial category based on the preset correspondence between each classification condition and the intent category and the pre-selected category.

[0126] Furthermore, the category determination module is specifically used to search for classification conditions corresponding to the pre-selected category from the preset correspondence between each classification condition and the intention category, and to determine the classification conditions satisfied by the target feedback content from the searched classification conditions, and based on the correspondence, determine the intention category corresponding to the classification conditions satisfied by the target feedback content as the initial category.

[0127] Furthermore, the vector generation module is specifically used to determine the key participles contained in the target feedback content, wherein the key participles are participles belonging to a preset participle type, and generate word vectors for the key participles, and based on the word vectors for the key participles, generate sentence vectors representing the target feedback content.

[0128] Furthermore, the intent classification model is: an intent classification model trained based on a random forest classification model.

[0129] Furthermore, the device also includes:

[0130] The subcategory determination module is used to determine the subcategory that has a mapping relationship with the preset segmentation contained in the target feedback content as the subcategory of the target feedback content according to the preset mapping relationship between the preset segmentation and the subcategory under the intention recognition result.

[0131] The embodiment of the present invention further provides an electronic device, such as Figure 4 As shown, it includes a processor 401, a communication interface 402, a memory 403 and a communication bus 404, wherein the processor 401, the communication interface 402, and the memory 403 communicate with each other through the communication bus 404.

[0132] Memory 403, used for storing computer programs;

[0133] The processor 401 is used to implement the above-mentioned method steps for identifying user intention for feedback content when executing the program stored in the memory 403.

[0134] The communication bus mentioned in the above electronic device can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0135] The communication interface is used for communication between the above electronic device and other devices.

[0136] The memory may include a random access memory (RAM) or a non-volatile memory (NVM), such as at least one disk memory. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.

[0137] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0138] In another embodiment of the present invention, a computer-readable storage medium is provided, in which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above-mentioned methods for identifying user intent for feedback content are implemented.

[0139] In another embodiment of the present invention, a computer program product including instructions is provided. When the computer program product is executed on a computer, the computer executes any one of the user intention recognition methods for feedback content in the above embodiments.

[0140] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented by software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from a website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state hard disk Solid State Disk (SSD)), etc.

[0141] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0142] Each embodiment in this specification is described in a related manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for devices, electronic devices, computer-readable storage media, and computer program products, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments.

[0143] The above description is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.

Claims

1. A method for identifying user intentions for feedback content, characterized in that: include: Obtain target feedback content to be identified; Based on the preset correspondence between each classification condition and the intention category, determine the intention category corresponding to the classification condition satisfied by the target feedback content as the initial category; wherein the intention category corresponding to each classification condition is the intention category that can be represented by the feedback content satisfying the classification condition; If it is determined that the initial category is a specified category, using a pre-trained intent classification model to determine an intent recognition result of the target feedback content; Otherwise, determining the initial category as the intention recognition result of the target feedback content; The designated category is an intent category in which a similar intent category exists; the intent classification model is a classification model trained based on a plurality of sample feedback contents, and the plurality of sample feedback contents include: sample feedback contents capable of representing the designated category, and sample feedback contents capable of representing intent categories similar to the designated category; Before determining, based on the preset correspondence between each classification condition and the intention category, the intention category corresponding to the classification condition satisfied by the target feedback content as the initial category, the method further includes: Generate a sentence vector representing the target feedback content as a target vector; Calculating the distance between the target vector and each cluster in the cluster space, wherein the cluster space is established based on the sentence vector of the sample feedback content, and each cluster in the cluster space is associated with an intent category; Determine a target cluster whose distance to the target vector is less than a preset threshold, and determine an intent category associated with the target cluster as a pre-selected category; The determining, based on the preset correspondence between each classification condition and the intention category, the intention category corresponding to the classification condition satisfied by the target feedback content as the initial category includes: Based on the preset correspondence between each classification condition and the intention category and the pre-selected category, determining the intention category corresponding to the classification condition satisfied by the target feedback content as the initial category; The determining, based on the preset correspondence between each classification condition and the intention category and the pre-selected category, the intention category corresponding to the classification condition satisfied by the target feedback content as the initial category includes: From the preset correspondence between each classification condition and the intended category, searching for the classification condition corresponding to the pre-selected category; The classification conditions satisfied by the target feedback content are determined from the searched classification conditions, and based on the corresponding relationship, the intention category corresponding to the classification conditions satisfied by the target feedback content is determined as the initial category.

2. The method according to claim 1, characterized in that The generating a sentence vector representing the target feedback content includes: Determining key participles contained in the target feedback content, wherein the key participles are participles belonging to a preset participle type; Generate a word vector for the key word segmentation, and generate a sentence vector representing the target feedback content based on the word vector for the key word segmentation.

3. The method according to any one of claims 1 to 2, characterized in that: The intent classification model is: an intent classification model trained based on a random forest classification model.

4. The method according to any one of claims 1 to 2, characterized in that: The method further comprises: According to a preset mapping relationship between a preset segmentation word and a subcategory under the intention recognition result, a subcategory having a mapping relationship with the preset segmentation word included in the target feedback content is determined as a subcategory of the target feedback content.

5. A device for identifying user intention for feedback content, characterized in that: include: A content acquisition module, used to acquire target feedback content to be identified; A category determination module, configured to determine, based on a preset correspondence between each classification condition and the intention category, the intention category corresponding to the classification condition satisfied by the target feedback content as the initial category; wherein the intention category corresponding to each classification condition is the intention category that can be represented by the feedback content satisfying the classification condition; A result determination module, for determining the intention recognition result of the target feedback content by using a pre-trained intention classification model if the initial category is determined to be a specified category; otherwise, determining the initial category as the intention recognition result of the target feedback content; The designated category is an intent category in which a similar intent category exists; the intent classification model is a classification model trained based on a plurality of sample feedback contents, and the plurality of sample feedback contents include: sample feedback contents capable of representing the designated category, and sample feedback contents capable of representing intent categories similar to the designated category; The device also includes: a vector generating module, configured to generate a sentence vector representing the target feedback content as a target vector before the category determining module determines the intent category corresponding to the classification condition satisfied by the target feedback content as an initial category based on the preset correspondence between each classification condition and the intent category; An example calculation module is provided, which is used to calculate the distance between the target vector and each cluster in the cluster space, wherein the cluster space is established based on the sentence vector of the sample feedback content, and each cluster in the cluster space is associated with an intent category; A cluster determination module, used to determine a target cluster whose distance to the target vector is less than a preset threshold, and determine an intent category associated with the target cluster as a pre-selected category; The category determination module is specifically configured to determine, based on a preset correspondence between each classification condition and the intention category and the preselected category, an intention category corresponding to the classification condition satisfied by the target feedback content as an initial category; The category determination module is specifically used to search for the classification conditions corresponding to the preselected category from the preset correspondence between each classification condition and the intention category, and determine the classification conditions satisfied by the target feedback content from the searched classification conditions, and based on the correspondence, determine the intention category corresponding to the classification conditions satisfied by the target feedback content as the initial category.

6. The device according to claim 5, characterized in that The vector generation module is specifically used to determine the key participles contained in the target feedback content, wherein the key participles are participles belonging to a preset participle type, and generate word vectors for the key participles, and based on the word vectors for the key participles, generate sentence vectors representing the target feedback content.

7. The device according to any one of claims 5 to 6, characterized in that: The intent classification model is: an intent classification model trained based on a random forest classification model.

8. The device according to any one of claims 5 to 6, characterized in that: The device also includes: A subcategory determination module is used to determine a subcategory that has a mapping relationship with the preset segmentation contained in the target feedback content as a subcategory of the target feedback content according to a preset mapping relationship between the preset segmentation and the subcategory under the intention recognition result.

9. An electronic device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; Memory, used to store computer programs; A processor, for implementing the method steps described in any one of claims 1 to 4 when executing a program stored in a memory.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method steps described in any one of claims 1 to 4 are implemented.

Images