Recommendation method, device and device for recommendation

Abstract

The embodiments of the present invention provide a recommendation method, device and device for recommendation. The method specifically includes: obtaining voice information input by a user; performing voice recognition on the voice information to obtain text information corresponding to the voice information; performing feature extraction on the voice information to obtain voice features corresponding to the voice information; determining personalized tags of the voice information based on the voice features, wherein the personalized tags include at least any one of the following: gender, emotion, health status, and region; determining recommended information based on the text information and the personalized tags. The embodiments of the present invention can improve the accuracy of recommended information.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a recommendation method, device and a device for recommendation. Background Art

[0002] With the continuous development of speech recognition technology, the application of recommendation based on speech recognition is becoming more and more widespread. Among them, speech recognition technology is a technology that takes speech as the research object, uses signal processing and recognition technology to allow the machine to automatically recognize and understand human spoken language, and then converts the speech signal into corresponding text or commands.

[0003] The recommendation system uses speech recognition technology to convert the voice data input by the user into text, and recommends relevant information to the user based on the content of the text. However, each sentence input by the user is usually a short text. Short texts have the characteristics of short length, sparse features, irregularity, and great ambiguity. Therefore, obtaining recommended information based on the recognized short text may result in inaccurate recommended information and not meet the user's intention. Summary of the invention

[0004] The embodiments of the present invention provide a recommendation method, a device and a device for recommendation, which can improve the accuracy of recommendation information.

[0005] In order to solve the above problem, an embodiment of the present invention discloses a recommendation method, which includes:

[0006] Get the voice information input by the user;

[0007] Performing voice recognition on the voice information to obtain text information corresponding to the voice information;

[0008] Extracting features from the voice information to obtain voice features corresponding to the voice information;

[0009] Determine, according to the voice feature, a personalized label for the voice information, wherein the personalized label includes at least any one of the following: gender, emotion, health condition, and region;

[0010] Determine recommended information according to the text information and the personalized tag.

[0011] On the other hand, an embodiment of the present invention discloses a recommendation device, the device comprising:

[0012] A voice acquisition module is used to acquire the voice information input by the user;

[0013] A speech recognition module, used to perform speech recognition on the speech information to obtain text information corresponding to the speech information;

[0014] A feature extraction module, used to extract features from the voice information to obtain voice features corresponding to the voice information;

[0015] A label determination module, used to determine a personalized label of the voice information according to the voice feature, wherein the personalized label includes at least any one of the following: gender, emotion, health condition, and region;

[0016] The information recommendation module is used to determine the recommended information according to the text information and the personalized tag.

[0017] In another aspect, an embodiment of the present invention discloses a device for recommendation, comprising a memory and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by one or more processors. The one or more programs include instructions for performing the following operations:

[0018] Get the voice information input by the user;

[0019] Performing voice recognition on the voice information to obtain text information corresponding to the voice information;

[0020] Extracting features from the voice information to obtain voice features corresponding to the voice information;

[0021] Determine, according to the voice feature, a personalized label for the voice information, wherein the personalized label includes at least any one of the following: gender, emotion, health condition, and region;

[0022] Determine recommended information according to the text information and the personalized tag.

[0023] On the other hand, an embodiment of the present invention discloses a machine-readable medium having instructions stored thereon, which, when executed by one or more processors, causes the device to perform one or more of the recommendation methods described above.

[0024] The embodiments of the present invention include the following advantages:

[0025] After acquiring the voice information input by the user, the embodiment of the present invention performs voice recognition on the voice information to obtain text information corresponding to the voice information; and performs feature extraction on the voice information to obtain the voice features corresponding to the voice information, and determines the personalized tags of the voice information based on the voice features; and then determines the recommended information based on the text information and the personalized tags. Among them, the personalized tags can be used to characterize the information of the user's own status contained in the voice features but cannot be expressed in the text information. For example, the personalized tags may include at least any one of the following: gender, emotion, health status, and region. Therefore, the embodiment of the present invention determines the recommended information based on the text information converted from the voice information and the personalized tags of the voice information, which is more in line with the user's own status and more in line with the user's intention, and can improve the accuracy of the recommended information. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0027] Figure 1 is a flowchart of steps of a recommended method embodiment of the present invention;

[0028] Figure 2 is a structural block diagram of a recommended device embodiment of the present invention;

[0029] Figure 3 is a block diagram of a device 800 for recommendation according to the present invention; and

[0030] Figure 4 It is a schematic diagram of the structure of the server in some embodiments of the present invention. DETAILED DESCRIPTION

[0031] 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 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.

[0032] Method Embodiment

[0033] Reference Figure 1 , shows a flowchart of a recommended method embodiment of the present invention, which may specifically include the following steps:

[0034] Step 101: Acquire voice information input by a user;

[0035] Step 102: Perform voice recognition on the voice information to obtain text information corresponding to the voice information;

[0036] Step 103: extracting features from the voice information to obtain voice features corresponding to the voice information;

[0037] Step 104: determining a personalized label of the voice information according to the voice feature, wherein the personalized label includes at least one of the following: gender, emotion, health condition, and region;

[0038] Step 105: Determine recommended information according to the text information and the personalized tag.

[0039] The recommendation method of the embodiment of the present invention can be applied to electronic devices, including but not limited to: servers, smart phones, tablet computers, e-book readers, MP3 (Moving Picture Experts Group Audio Layer III) players, MP4 (Moving Picture Experts Group Audio Layer IV) players, laptop computers, car computers, desktop computers, set-top boxes, smart TVs, wearable devices, etc.

[0040] The recommendation method of the embodiment of the present invention can be used to provide the user with recommendation information that is more in line with the user's intention based on the voice information input by the user. The voice information refers to a continuous voice, such as a sentence, a paragraph, etc. It can be understood that the embodiment of the present invention does not limit the source of the voice information. For example, the voice information can be a voice segment collected in real time by the recording function of the electronic device.

[0041] In an optional embodiment of the present invention, the acquiring of voice information input by the user may specifically include: acquiring voice information input, sent or received by the user through an instant messaging application.

[0042] The instant messaging application is an application that uses instant messaging technology to achieve online chatting and communication. The voice information obtained by the embodiment of the present invention may include: voice information input by the user through the instant messaging application, voice information sent by the user to the communication peer through the instant messaging application, and voice information received by the user from the communication peer through the instant messaging application. Therefore, the embodiment of the present invention can recommend recommended information that meets the user's intention in real time based on the voice information input, sent or received by the user.

[0043] After obtaining the voice information input by the user, the voice information can be preprocessed to eliminate the influence of aliasing, high-order harmonic distortion, high frequency and other factors caused by the vocal organs and acquisition equipment, and improve the quality of the voice information. The preprocessing can include pre-emphasis, framing, windowing and other operations.

[0044] After preprocessing the voice information, voice recognition can be performed on the preprocessed voice information to obtain text information corresponding to the voice information. In order to improve the accuracy of the recommended information, in addition to obtaining the text information corresponding to the voice information, the embodiment of the present invention can also perform feature extraction on the preprocessed voice information frame by frame to reduce the influence of information irrelevant to recognition in the voice information, reduce the amount of data required to be processed in the subsequent recognition stage, and generate voice features that characterize the speaker information in the voice information. The voice features may include MFCC (Mel-frequency cepstrum coefficient), or LPCC (Linear Prediction Cepstrum Coefficient), or filter bank feature, etc.

[0045] The extracted voice features contain important information about the user's own state that cannot be expressed by text information (referred to as personalized tags in the embodiment of the present invention), such as the user's gender, emotion, physical condition, etc. The embodiment of the present invention can determine the personalized tags of the voice information based on the voice features, which are used to express the characteristic information of the user's own state contained in the voice features, and the personalized tags can include at least any one of the following: gender, emotion, health condition, and region.

[0046] In an application example of the present invention, it is assumed that the voice information input by the user is "My throat is uncomfortable now, cough cough cough...". The embodiment of the present invention can perform voice recognition on the voice information to obtain corresponding text information, and can also perform feature extraction on the voice information to obtain voice features corresponding to the voice information, and determine the personalized label of the voice information based on the voice features. Since the voice information contains a severe cough, the personalized label can be determined as "cold" based on the extracted voice features, and the recommended information can be determined based on the text information "My throat is uncomfortable now, cough cough cough..." obtained by voice recognition and the personalized label "cold".

[0047] For example, based on the text information, it can be known that the user has a sore throat, and based on the personalized tag, it can be known that the user's current state is a cold. Therefore, medical information and drug information for treating colds and sore throats can be obtained as recommended information and recommended to the user.

[0048] The embodiment of the present invention determines the recommended information based on the text information converted from the voice information and in combination with the personalized tag of the voice information. Since the personalized tag is determined based on the voice features extracted from the voice information, it can express important information about the user's own state that cannot be expressed by the text information. Therefore, the recommended information obtained by combining the personalized tag is more in line with the user's own state, and further more in line with the user's intention, which can improve the accuracy of the recommended information.

[0049] In an optional embodiment of the present invention, determining the personalized tag of the voice information according to the voice feature in step 104 may specifically include:

[0050] The speech feature is input into a classification model so as to output a personalized label corresponding to the speech feature through the classification model; wherein the classification model is trained based on speech sample data that has been labeled with personalized labels.

[0051] The embodiment of the present invention can pre-train a classification model for identifying personalized labels, and input the speech features extracted from the speech information into the classification model to output the personalized labels corresponding to the speech information. The classification model analyzes and processes the speech features in combination with the classification algorithm to obtain a classification result. It can be understood that the embodiment of the present invention does not limit the type of the classification algorithm. The classification algorithm can include support vector machines, multi-layer perceptrons, recurrent neural networks, convolutional neural networks, and probabilistic linear discriminant analysis, etc.

[0052] In an optional embodiment of the present invention, before inputting the speech features into the classification model, the method may further include:

[0053] Step S11, obtaining voice sample data, wherein the voice sample data is annotated with a corresponding personalized label;

[0054] Step S12: extracting features from the speech sample data to obtain speech sample features corresponding to the speech sample data;

[0055] Step S13, inputting the speech sample features into an initial classification model for training;

[0056] Step S14: During the training process, the model parameters of the initial classification model are adjusted until the personalized label output by the initial classification model matches the personalized label annotated with the speech sample data, thereby obtaining a trained classification model.

[0057] The embodiment of the present invention can collect a large amount of voice sample data, and annotate the voice sample data with corresponding personalized tags for training a classification model.

[0058] It should be noted that the embodiments of the present invention can classify the voice sample data according to the marked personalized labels, and train different classification models for the voice sample data of different personalized label categories. For example, different classification models are trained for different categories of personalized labels such as the user's emotions, health status, gender, etc. For example, the emotion classification model can be used to output the user's emotions, such as happiness, anxiety, excitement, etc.; the health status classification model can be used to output the user's health status, such as cold, fever, headache, etc.; the gender classification model can be used to output the user's gender, such as male and female; the age classification model can be used to output the user's age group, such as the elderly, middle-aged, young, and children. When applied, the voice features can be input into the classification models of different personalized label categories to obtain the classification results of each classification model. Alternatively, the embodiments of the present invention can also train a general classification model based on the voice sample data. For example, the voice sample data used to train the classification model includes two personalized labels, health status and gender, and the trained classification model can output two classification results of health status and gender.

[0059] After collecting the speech sample data, feature extraction can be performed on the speech sample data to obtain speech sample features corresponding to the speech sample data; then the speech sample features are input into the initial classification model for training, and during the training process, the model parameters of the initial classification model are adjusted until the personalized label output by the initial classification model matches the personalized label annotated with the speech sample data, thereby obtaining a trained classification model.

[0060] It is understandable that the embodiment of the present invention does not limit the specific type of the classification model, and the classification model may include DNN (Deep Neural Networks). The deep neural network may fuse multiple neural networks, including but not limited to at least one or a combination, superposition, and nesting of at least two of the following: CNN (Convolutional Neural Network), LSTM (Long Short-Term Memory) network, RNN (Simple Recurrent Neural Network), attention neural network, etc.

[0061] In an optional embodiment of the present invention, step 105 determines the recommended information according to the text information and the personalized tag, which may specifically include:

[0062] Step S21, retrieving candidate information according to entity words in the text information;

[0063] Step S22: Determine recommended information according to the personalized tag and the candidate information.

[0064] Among them, entity words are nouns or pronouns with specific meanings in the text, and entity words can include names of people, places, institutions, proper nouns, etc. For example, entity words can include: names of home appliances, clothing, brand names, drug names, service names (such as car washing, home appliance repair, housekeeping information, etc.), ticket names, disease names, etc.

[0065] The embodiment of the present invention can identify entity words in text information, input the identified entity words as query words into a search engine, and retrieve candidate information. For example, if the text information is identified to contain the entity word "movie", the entity word can be used as a query word to search, such as retrieving information related to recently released movies, information related to discounted movie tickets, etc., and the above search results are used as candidate information.

[0066] Next, the candidate information can be screened according to the personalized tags to determine the recommended information. The personalized tags can include at least any one of the following: gender, emotion, health status, and region. For example, combined with the personalized tags of gender, sports news and other related information can be recommended to male users, and cosmetics and other related information can be recommended to female users. Combined with the personalized tags of the region, relevant information such as goods or services (such as real estate, restaurants, etc.) can be recommended to users in a certain region in a targeted manner.

[0067] Still taking the example of the entity word "movie" contained in the above-mentioned recognized text information as an example, assuming that based on the voice information input by the user, the personalized label corresponding to the user's emotion is determined to be "happy", indicating that the user's emotional state is relatively good, then the candidate information retrieved in step S21 can be further screened out, suitable for watching when the emotional state is relatively good, as recommended information.

[0068] In another application example of the present invention, the personalized tag can also indicate the region to which the user belongs. For example, the embodiment of the present invention can determine the user's accent based on the voice features in the voice information input by the user, and then determine the region to which the user belongs. For example, based on the accent features in the user's voice, it is determined that the region to which the user belongs is "Shanghai", and then recommended information that conforms to the habits of Shanghainese can be recommended to the user. For example, if the voice information input by the user contains entity words about food, then in combination with the personalized tag of the user, food information that is more in line with the tastes of Shanghainese can be used as recommended information.

[0069] Therefore, the embodiment of the present invention converts the voice information input by the user into text information, and extracts voice features in the voice information, determines a personalized label for characterizing the user's current state based on the voice features, and jointly determines the recommended information by combining the text information and the voice features, so that the determined recommended information not only conforms to the user's text expression, but also conforms to the user's current state, which can improve the accuracy of information recommendation.

[0070] In an optional embodiment of the present invention, after obtaining the text information corresponding to the voice information in step 102, the method may further include: performing intent recognition on the text information to determine the user's recommendation intent.

[0071] In order to further improve the accuracy of information recommendation and make the recommended information more in line with the user's intention, after performing voice recognition on the voice information input by the user and converting it into text information, the embodiment of the present invention can also perform intent recognition on the text information to determine the user's recommendation intention. Specifically, the text information can be segmented to obtain keywords in the text information, and the recommendation intention that meets the user can be determined based on the keywords.

[0072] For example, the voice information input by the user is converted into the following text information: "My throat is uncomfortable now, cough cough cough...", and the text information is segmented, and it can be obtained that the text information includes the following keywords: "I", "throat", "uncomfortable". According to the above keywords, it can be determined that the user's recommendation intention is related to medical health. For another example, for the text information "Do you want to eat hot pot tonight?", the text information includes the keywords "tonight" and "eat hot pot". According to the above keywords, it can be determined that the user's recommendation intention is related to restaurants.

[0073] The recognition of recommendation intent can be regarded as a problem of classifying semantic expressions, and classification methods can be used for recognition, including support vector machines and deep neural networks. It can be understood that the embodiments of the present invention do not limit the specific method of identifying recommendation intent. For example, a large number of text samples can be collected in advance, the recommendation intent of the text samples can be annotated, and a recognition model for identifying recommendation intent can be obtained by training. In a specific application, the text information converted from voice information is input into the recognition model, and the corresponding recommendation intent can be output.

[0074] After determining the user's recommendation intention, the recommendation intention, personalized tags, and voice features can be combined to jointly determine the recommendation information to further improve the accuracy of the recommendation information.

[0075] Among them, personalized tags can supplement the user's recommendation intention, such as reflecting the user's emotions, mood, health status, etc., so that the recommended information is more in line with the user's current state. In addition, in the process of determining the recommended information, the embodiment of the present invention can further combine voice features, which are extracted from the user's original voice data, so that the final recommendation result not only depends on the analysis result obtained by natural language understanding, but also can reflect the most original voice features, so that the recommendation result depends on a variety of features with richer representations, further improving the accuracy of the recommendation result.

[0076] Optionally, determining the recommended information according to the text information and the personalized tag may specifically include:

[0077] Step S31, generating a first vector corresponding to the recommendation intention, a second vector corresponding to the personalized tag, and a third vector corresponding to the voice feature;

[0078] Step S32: concatenate the first vector, the second vector, and the third vector to obtain a user feature vector;

[0079] Step S33, generating a fourth vector according to the candidate information corresponding to the text information;

[0080] Step S34: input the user feature vector and the fourth vector into a recommendation model, so as to output recommendation information through the recommendation model.

[0081] The embodiments of the present invention can use vectorization technology to vectorize recommendation intentions, personalized tags, voice features, and recommendation information (such as web page information), and determine the recommendation information through technologies such as nearest neighbor or deep learning-based recommendation models.

[0082] For recommendation intent and personalized tags, a word vector model can be used for vector representation. Specifically, the recommendation intent is input into the word vector model to obtain a first vector; the personalized tag is input into the word vector model to obtain a second vector. The word vector model may include Word2Vec or GloVe, etc.

[0083] For speech features, the corresponding third vector may be a Mel-Frequency Cepstral Coefficient (MFCC) feature vector. Specifically, the preprocessed speech information may be subjected to Fourier transform, and the transformed signal may be further subjected to filtering and DCT (Discrete Cosine Transform), and finally the MFCC feature vector corresponding to the speech information may be obtained as the third vector.

[0084] The user feature vector can be obtained by concatenating the first vector, the second vector, and the third vector. The embodiment of the present invention can calculate the correlation between the user feature vector and the vector of the recommended information, and select the recommended information with higher correlation to recommend to the user.

[0085] Specifically, candidate information to be recommended (such as web page information) can be determined based on the text information, and a fourth vector can be obtained by vectorizing the candidate information. The user feature vector and the fourth vector are input into a pre-trained recommendation model. The recommendation model can calculate the correlation between the user feature vector and the fourth vector, sort the candidate information according to the correlation, and select the top n (n is a positive integer) candidate information as recommended information for recommendation.

[0086] It can be understood that the embodiment of the present invention does not limit the structure and training method of the recommendation model.

[0087] The recommendation model may be a deep neural network model, and the deep neural network may integrate multiple neural networks, including but not limited to a combination, superposition, and nesting of at least one or at least two of the following: CNN network, RNN, attention neural network, etc.

[0088] In an example of the present invention, the recommendation model can be trained by the following steps: first, historical recommendation data corresponding to the historical voice information of a large number of users can be collected, and the historical recommendation data can be annotated according to whether the user has clicked on the historical recommendation data to obtain the annotated results. Then, the historical recommendation intention is determined according to the text information corresponding to the historical voice information; the historical personalized label is determined according to the voice features corresponding to the historical voice information. Next, a historical first vector corresponding to the historical recommendation intention, a historical second vector corresponding to the historical personalized label, a historical third vector corresponding to the voice features of the historical voice information, and a historical fourth vector corresponding to the historical candidate information are generated. Finally, the historical user feature vector and the historical fourth vector obtained by splicing the historical first vector, the historical second vector, and the historical third vector are input into the initial recommendation model, and the initial recommendation model is optimized according to the difference between the recommendation result output by the initial recommendation model and the annotated result, until the difference between the recommendation result output by the initial recommendation model and the annotated result is less than the preset threshold, and the trained recommendation model is obtained.

[0089] In an optional embodiment of the present invention, after determining the recommendation intention of the user, the method may further include:

[0090] Step S41, obtaining slot information corresponding to the recommendation intention;

[0091] Step S42: fill the slots of the text information according to the slot information to obtain the slot values ​​corresponding to the slot information in the text information.

[0092] In an embodiment of the present invention, slot information corresponding to the recommendation intent may be preset. The slot information refers to the slot required in the vertical search field, and different recommendation intents may correspond to different slot information. For example, for shopping intent, the corresponding slot information may include: product name, brand name, model, price, etc. For another example, for medical and health intent, the corresponding slot information may include: body part, physical condition, person, time, etc.

[0093] If the user's recommendation intention is determined to be a medical health intention based on the text information, the slot information corresponding to the pre-set medical health intention can be obtained. Then, based on the slot information, the text information is slot-filled to obtain the slot values ​​corresponding to the slot information in the text information. Among them, slot filling can be regarded as a sequence labeling problem, which is to match each word in the text information with the pre-set slot information and identify the slot values ​​corresponding to the slot information in the text information.

[0094] The generating the first vector corresponding to the recommendation intent may specifically include: generating the first vector corresponding to the recommendation intent according to each slot value corresponding to the slot information in the text information.

[0095] Since the slot value in the text information matches the slot information of the recommended intent corresponding to the text information, after determining the slot values ​​corresponding to the slot information in the text information, each slot value in the text information can be vectorized to obtain the first vector corresponding to the recommended intent.

[0096] In an application example of the present invention, it is assumed that the voice information input by the user is "My throat is uncomfortable now, cough cough cough...". First, the voice information is subjected to voice recognition to obtain the corresponding text information. According to the keywords in the text information (such as: "I", "throat", "uncomfortable"), it can be determined that the user's recommendation intention is a medical health intention. Therefore, the slot information corresponding to the medical health intention can be obtained (such as: body part, physical condition, person, time), and the slot values ​​corresponding to the slot information in the text information are obtained by slot filling as follows: body part = "throat", physical condition = "uncomfortable", person = "I", time = "now". In addition, the voice information can be feature extracted to obtain the voice features corresponding to the voice information, and the personalized label of the voice information can be determined based on the voice features. Since the voice information contains a severe cough, the personalized label can be determined as "cold" based on the extracted voice features. Next, according to the values ​​of each slot in the text information, the first vector corresponding to the recommendation intention can be generated. Specifically, (body part = "throat", physical condition = "uncomfortable", person = "I", time = "now") can be vectorized to obtain the first vector; the personalized label "cold" can be vectorized to obtain the second vector; and the voice features extracted from the user input voice information "I have a sore throat now, cough cough cough..." can be vectorized to obtain the third vector. Finally, the first vector, the second vector and the third vector are concatenated to obtain the user feature vector, and the fourth vector corresponding to the user feature vector and the candidate information is input into the recommendation model to obtain the recommended information.

[0097] After determining the recommended information according to the voice information input by the user, the recommended information can be displayed to the user. Specifically, the recommended information can be displayed to the user in the chat interface of the instant messaging application currently used by the user, so that the user can directly view the detailed content of the recommended information by clicking on the recommended information, thereby improving the efficiency and convenience of the user in obtaining the recommended information content.

[0098] Optionally, the embodiment of the present invention may also send the recommendation information to the user's communication peer. Since the current user and the communication peer user are chatting using an instant messaging application, both may have the same recommendation intention for the same current topic. Therefore, the current user's recommendation information may also be recommended to the communication peer user, so that the communication peer user can quickly obtain the required recommendation information.

[0099] In summary, after acquiring the voice information input by the user, the embodiment of the present invention performs voice recognition on the voice information to obtain text information corresponding to the voice information; and performs feature extraction on the voice information to obtain the voice features corresponding to the voice information, and determines the personalized tags of the voice information based on the voice features; and then determines the recommended information based on the text information and the personalized tags. Among them, the personalized tags can be used to characterize the information of the user's own status contained in the voice features but cannot be expressed in the text information. For example, the personalized tags may include at least any one of the following: gender, emotion, health status, and region. Therefore, the embodiment of the present invention determines the recommended information based on the text information converted from the voice information and the personalized tags of the voice information, which is more in line with the user's own status and more in line with the user's intention, and can improve the accuracy of the recommended information.

[0100] It should be noted that, for the sake of simplicity, the method embodiments are described as a series of action combinations, but those skilled in the art should be aware that the embodiments of the present invention are not limited by the order of the actions described, because according to the embodiments of the present invention, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily required by the embodiments of the present invention.

[0101] Device Embodiment

[0102] Reference Figure 2 , shows a structural block diagram of a recommended device embodiment of the present invention, and the device may specifically include:

[0103] The voice acquisition module 201 is used to acquire the voice information input by the user;

[0104] A speech recognition module 202, configured to perform speech recognition on the speech information to obtain text information corresponding to the speech information;

[0105] A feature extraction module 203 is used to extract features from the voice information to obtain voice features corresponding to the voice information;

[0106] The label determination module 204 is used to determine the personalized label of the voice information according to the voice feature, and the personalized label includes at least any one of the following: gender, emotion, health status, and region;

[0107] The information recommendation module 205 is used to determine recommended information according to the text information and the personalized tag.

[0108] Optionally, the information recommendation module 205 may specifically include:

[0109] A retrieval submodule, used to retrieve candidate information based on entity words in the text information;

[0110] The determination submodule is used to determine the recommended information according to the personalized tag and the candidate information.

[0111] Optionally, the device may further include:

[0112] An intention determination module, used to identify the intention of the text information and determine the recommendation intention of the user;

[0113] The information recommendation module includes:

[0114] A first generating submodule, used to generate a first vector corresponding to the recommendation intention, a second vector corresponding to the personalized tag, and a third vector corresponding to the voice feature;

[0115] A second generating submodule, configured to concatenate the first vector, the second vector, and the third vector to obtain a user feature vector;

[0116] A third generating submodule, used for generating a fourth vector according to the candidate information corresponding to the text information;

[0117] The information determination submodule is used to input the user feature vector and the fourth vector into a recommendation model to output recommendation information through the recommendation model.

[0118] Optionally, the device may further include:

[0119] A slot acquisition module, used to acquire slot information corresponding to the recommendation intent;

[0120] A slot filling module, used to fill the slots of the text information according to the slot information, and obtain each slot value corresponding to the slot information in the text information;

[0121] The first vector determination submodule is specifically configured to generate a first vector corresponding to the recommendation intent according to each slot value corresponding to the slot information in the text information.

[0122] Optionally, the label determination module is specifically used to input the speech feature into a classification model to output a personalized label corresponding to the speech feature through the classification model; wherein the classification model is trained based on speech sample data that has been labeled with personalized labels.

[0123] Optionally, the device may further include:

[0124] A sample acquisition module, used to acquire voice sample data, wherein the voice sample data is annotated with a corresponding personalized label;

[0125] A feature acquisition module, used to extract features from the speech sample data to obtain speech sample features corresponding to the speech sample data;

[0126] A feature training module, used for inputting the speech sample features into an initial classification model for training;

[0127] The parameter adjustment module is used to adjust the model parameters of the initial classification model during the training process until the personalized label output by the initial classification model matches the personalized label annotated by the speech sample data, thereby obtaining a trained classification model.

[0128] Optionally, the voice acquisition module is specifically used to acquire voice information input, sent or received by a user through an instant messaging application.

[0129] As for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0130] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0131] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0132] An embodiment of the present invention provides a device for recommendation, including a memory and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by one or more processors, wherein the one or more programs include instructions for performing the following operations: obtaining voice information input by a user; performing voice recognition on the voice information to obtain text information corresponding to the voice information; performing feature extraction on the voice information to obtain voice features corresponding to the voice information; determining personalized tags for the voice information based on the voice features, wherein the personalized tags include at least any one of the following: gender, emotion, health status, and region; and determining recommendation information based on the text information and the personalized tags.

[0133] Figure 3 8 is a block diagram of a device 800 for recommendation according to an exemplary embodiment. For example, the device 800 may be a mobile phone, a computer, a digital broadcast terminal, a messaging device, a game console, a tablet device, a medical device, a fitness device, a personal digital assistant, etc.

[0134] Reference Figure 3 , the device 800 may include one or more of the following components: a processing component 802 , a memory 804 , a power component 806 , a multimedia component 808 , an audio component 810 , an input / output (I / O) interface 812 , a sensor component 814 , and a communication component 816 .

[0135] The processing component 802 generally controls the overall operation of the device 800, such as operations associated with display, phone calls, data communications, camera operations, and recording operations. The processing component 802 may include one or more processors 820 to execute instructions to complete all or part of the steps of the above-mentioned method. In addition, the processing component 802 may include one or more modules to facilitate the interaction between the processing component 802 and other components. For example, the processing component 802 may include a multimedia module to facilitate the interaction between the multimedia component 808 and the processing component 802.

[0136] The memory 804 is configured to store various types of data to support operations on the device 800. Examples of such data include instructions for any application or method operating on the device 800, contact data, phone book data, messages, pictures, videos, etc. The memory 804 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0137] The power supply component 806 provides power to the various components of the device 800. The power supply component 806 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power for the device 800.

[0138] The multimedia component 808 includes a screen that provides an output interface between the device 800 and the user. In some embodiments, the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touch, slide, and gestures on the touch panel. The touch sensor may not only sense the boundaries of the touch or slide action, but also detect the duration and pressure associated with the touch or slide operation. In some embodiments, the multimedia component 808 includes a front camera and / or a rear camera. When the device 800 is in an operating mode, such as a shooting mode or a video mode, the front camera and / or the rear camera may receive external multimedia data. Each front camera and rear camera may be a fixed optical lens system or have a focal length and optical zoom capability.

[0139] The audio component 810 is configured to output and / or input audio signals. For example, the audio component 810 includes a microphone (MIC), and when the device 800 is in an operating mode, such as a call mode, a recording mode, and a voice information processing mode, the microphone is configured to receive an external audio signal. The received audio signal can be further stored in the memory 804 or sent via the communication component 816. In some embodiments, the audio component 810 also includes a speaker for outputting audio signals.

[0140] I / O interface 812 provides an interface between processing component 802 and peripheral interface modules, such as keyboards, click wheels, buttons, etc. These buttons may include but are not limited to: home button, volume button, start button, and lock button.

[0141] The sensor assembly 814 includes one or more sensors for providing various aspects of status assessment for the device 800. For example, the sensor assembly 814 can detect the open / closed state of the device 800, the relative positioning of components, such as the display and keypad of the device 800, and the sensor assembly 814 can also detect the position change of the device 800 or a component of the device 800, the presence or absence of user contact with the device 800, the orientation or acceleration / deceleration of the device 800, and the temperature change of the device 800. The sensor assembly 814 may include a proximity sensor configured to detect the presence of a nearby object without any physical contact. The sensor assembly 814 may also include an optical sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, the sensor assembly 814 may also include an accelerometer, a gyroscope sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.

[0142] The communication component 816 is configured to facilitate wired or wireless communication between the device 800 and other devices. The device 800 can access a wireless network based on a communication standard, such as WiFi, 2G or 3G, or a combination thereof. In an exemplary embodiment, the communication component 816 receives a broadcast signal or broadcast-related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component 816 also includes a near field communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on radio frequency information processing (RFID) technology, infrared data association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology and other technologies.

[0143] In an exemplary embodiment, the apparatus 800 may be implemented by one or more application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components to perform the above method.

[0144] In an exemplary embodiment, a non-transitory computer-readable storage medium including instructions is also provided, such as a memory 804 including instructions, and the instructions can be executed by the processor 820 of the device 800 to perform the above method. For example, the non-transitory computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc.

[0145] Figure 41 is a schematic diagram of the structure of the server in some embodiments of the present invention. The server 1900 may have relatively large differences due to different configurations or performances, and may include one or more central processing units (CPU) 1922 (for example, one or more processors) and memory 1932, and one or more storage media 1930 (for example, one or more mass storage devices) storing application programs 1942 or data 1944. Among them, the memory 1932 and the storage medium 1930 can be short-term storage or permanent storage. The program stored in the storage medium 1930 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the server. Furthermore, the central processing unit 1922 can be configured to communicate with the storage medium 1930 and execute a series of instruction operations in the storage medium 1930 on the server 1900.

[0146] The server 1900 may also include one or more power supplies 1926, one or more wired or wireless network interfaces 1950, one or more input and output interfaces 1958, one or more keyboards 1956, and / or, one or more operating systems 1941, such as Windows ServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM, etc.

[0147] A non-transitory computer-readable storage medium, when the instructions in the storage medium are executed by a processor of a device (server or terminal), enables the device to perform Figure 1 The recommended method is shown.

[0148] A non-temporary computer-readable storage medium, when the instructions in the storage medium are executed by a processor of a device (server or terminal), the device is able to execute a recommendation method, the method comprising: obtaining voice information input by a user; performing voice recognition on the voice information to obtain text information corresponding to the voice information; performing feature extraction on the voice information to obtain voice features corresponding to the voice information; determining personalized tags for the voice information based on the voice features, the personalized tags including at least any one of the following: gender, emotion, health status, and region; determining recommended information based on the text information and the personalized tags.

[0149] The embodiment of the present invention discloses A1, a recommendation method, comprising:

[0150] Get the voice information input by the user;

[0151] Performing voice recognition on the voice information to obtain text information corresponding to the voice information;

[0152] Extracting features from the voice information to obtain voice features corresponding to the voice information;

[0153] Determine, according to the voice feature, a personalized label for the voice information, wherein the personalized label includes at least any one of the following: gender, emotion, health condition, and region;

[0154] Determine recommended information according to the text information and the personalized tag.

[0155] A2. The method according to A1, wherein determining the recommended information according to the text information and the personalized tag comprises:

[0156] Retrieving candidate information according to entity words in the text information;

[0157] Determine recommended information according to the personalized tag and the candidate information.

[0158] A3. According to the method described in A1, after obtaining the text information corresponding to the voice information, the method further includes:

[0159] Performing intent recognition on the text information to determine the user's recommendation intent;

[0160] The determining of the recommended information according to the text information and the personalized tag includes:

[0161] Generate a first vector corresponding to the recommendation intention, a second vector corresponding to the personalized tag, and a third vector corresponding to the voice feature;

[0162] Concatenate the first vector, the second vector, and the third vector to obtain a user feature vector;

[0163] generating a fourth vector according to the candidate information corresponding to the text information;

[0164] The user feature vector and the fourth vector are input into a recommendation model to output recommendation information through the recommendation model.

[0165] A4. According to the method of A3, after determining the recommendation intention of the user, the method further includes:

[0166] Obtaining slot information corresponding to the recommendation intent;

[0167] Filling the slots of the text information according to the slot information to obtain the slot values ​​of each slot in the text information corresponding to the slot information;

[0168] The generating a first vector corresponding to the recommendation intention includes:

[0169] A first vector corresponding to the recommendation intent is generated according to each slot value corresponding to the slot information in the text information.

[0170] A5. The method according to A1, wherein determining the personalized label of the voice information according to the voice feature comprises:

[0171] The speech feature is input into a classification model so as to output a personalized label corresponding to the speech feature through the classification model; wherein the classification model is trained based on speech sample data that has been labeled with personalized labels.

[0172] A6. The method according to A5, before inputting the speech features into the classification model, the method further comprises:

[0173] Acquire voice sample data, wherein the voice sample data is annotated with a corresponding personalized label;

[0174] Extracting features from the speech sample data to obtain speech sample features corresponding to the speech sample data;

[0175] Inputting the speech sample features into an initial classification model for training;

[0176] During the training process, the model parameters of the initial classification model are adjusted until the personalized label output by the initial classification model matches the personalized label annotated with the speech sample data, thereby obtaining a trained classification model.

[0177] A7. According to any one of the methods A1 to A6, the step of obtaining voice information input by the user comprises:

[0178] Get the voice messages entered, sent, or received by the user through the instant messaging application.

[0179] The embodiment of the present invention discloses B8, a recommendation device, comprising:

[0180] A voice acquisition module is used to acquire the voice information input by the user;

[0181] A speech recognition module, used to perform speech recognition on the speech information to obtain text information corresponding to the speech information;

[0182] A feature extraction module, used to extract features from the voice information to obtain voice features corresponding to the voice information;

[0183] A label determination module, used to determine a personalized label of the voice information according to the voice feature, wherein the personalized label includes at least any one of the following: gender, emotion, health condition, and region;

[0184] The information recommendation module is used to determine the recommended information according to the text information and the personalized tag.

[0185] B9. According to the device described in B8, the information recommendation module includes:

[0186] A retrieval submodule, used to retrieve candidate information based on entity words in the text information;

[0187] The determination submodule is used to determine the recommended information according to the personalized tag and the candidate information.

[0188] B10. The device according to B8, further comprising:

[0189] An intention determination module, used to identify the intention of the text information and determine the recommendation intention of the user;

[0190] The information recommendation module includes:

[0191] A first generating submodule, used to generate a first vector corresponding to the recommendation intention, a second vector corresponding to the personalized tag, and a third vector corresponding to the voice feature;

[0192] A second generating submodule, configured to concatenate the first vector, the second vector, and the third vector to obtain a user feature vector;

[0193] A third generating submodule, used for generating a fourth vector according to the candidate information corresponding to the text information;

[0194] The information determination submodule is used to input the user feature vector and the fourth vector into a recommendation model to output recommendation information through the recommendation model.

[0195] B11. The device according to B10, further comprising:

[0196] A slot acquisition module, used to acquire slot information corresponding to the recommendation intent;

[0197] A slot filling module, used to fill the slots of the text information according to the slot information, and obtain each slot value corresponding to the slot information in the text information;

[0198] The first vector determination submodule is specifically configured to generate a first vector corresponding to the recommendation intent according to each slot value corresponding to the slot information in the text information.

[0199] B12. According to the device described in B8, the label determination module is specifically used to input the speech feature into a classification model to output a personalized label corresponding to the speech feature through the classification model; wherein the classification model is trained based on speech sample data that has been marked with personalized labels.

[0200] B13. The method according to B12, wherein the device further comprises:

[0201] A sample acquisition module, used to acquire voice sample data, wherein the voice sample data is annotated with a corresponding personalized label;

[0202] A feature acquisition module, used to extract features from the speech sample data to obtain speech sample features corresponding to the speech sample data;

[0203] A feature training module, used for inputting the speech sample features into an initial classification model for training;

[0204] The parameter adjustment module is used to adjust the model parameters of the initial classification model during the training process until the personalized label output by the initial classification model matches the personalized label annotated by the speech sample data, thereby obtaining a trained classification model.

[0205] B14. According to any one of the devices described in B8 to B13, the voice acquisition module is specifically used to acquire voice information input, sent or received by the user through an instant messaging application.

[0206] The embodiment of the present invention discloses C15, a device for recommendation, comprising a memory and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by one or more processors. The one or more programs include instructions for performing the following operations:

[0207] Get the voice information input by the user;

[0208] Performing voice recognition on the voice information to obtain text information corresponding to the voice information;

[0209] Extracting features from the voice information to obtain voice features corresponding to the voice information;

[0210] Determine, according to the voice feature, a personalized label for the voice information, wherein the personalized label includes at least any one of the following: gender, emotion, health condition, and region;

[0211] Determine recommended information according to the text information and the personalized tag.

[0212] C16. According to the apparatus of C15, determining the recommended information according to the text information and the personalized tag comprises:

[0213] Retrieving candidate information according to entity words in the text information;

[0214] Determine recommended information according to the personalized tag and the candidate information.

[0215] C17. The device according to C15, wherein the device is further configured to execute, by one or more processors, the one or more programs including instructions for performing the following operations:

[0216] Performing intent recognition on the text information to determine the recommendation intent of the user;

[0217] The determining of the recommended information according to the text information and the personalized tag includes:

[0218] Generate a first vector corresponding to the recommendation intention, a second vector corresponding to the personalized tag, and a third vector corresponding to the voice feature;

[0219] Concatenate the first vector, the second vector, and the third vector to obtain a user feature vector;

[0220] generating a fourth vector according to the candidate information corresponding to the text information;

[0221] The user feature vector and the fourth vector are input into a recommendation model to output recommendation information through the recommendation model.

[0222] C18. The device according to C17, wherein the device is further configured to execute, by one or more processors, the one or more programs including instructions for performing the following operations:

[0223] Obtaining slot information corresponding to the recommendation intent;

[0224] According to the slot information, the text information is slot-filled to obtain each slot value corresponding to the slot information in the text information;

[0225] The generating a first vector corresponding to the recommendation intention includes:

[0226] A first vector corresponding to the recommendation intent is generated according to each slot value corresponding to the slot information in the text information.

[0227] C19. According to the device described in C15, determining the personalized label of the voice information according to the voice feature includes:

[0228] The speech feature is input into a classification model so as to output a personalized label corresponding to the speech feature through the classification model; wherein the classification model is trained based on speech sample data that has been labeled with personalized labels.

[0229] C20. The device according to C19, wherein the device is further configured to execute, by one or more processors, the one or more programs including instructions for performing the following operations:

[0230] Acquire voice sample data, wherein the voice sample data is annotated with a corresponding personalized label;

[0231] Extracting features from the speech sample data to obtain speech sample features corresponding to the speech sample data;

[0232] Inputting the speech sample features into an initial classification model for training;

[0233] During the training process, the model parameters of the initial classification model are adjusted until the personalized label output by the initial classification model matches the personalized label annotated with the speech sample data, thereby obtaining a trained classification model.

[0234] C21. According to any one of C15 to C20, the step of obtaining voice information input by a user comprises:

[0235] Get the voice messages entered, sent, or received by the user through the instant messaging application.

[0236] An embodiment of the present invention discloses D22, a machine-readable medium having instructions stored thereon, which, when executed by one or more processors, causes the device to execute the recommended method as described in one or more of A1 to A7.

[0237] Those skilled in the art will readily appreciate other embodiments of the invention after considering the specification and practicing the invention disclosed herein. The present invention is intended to cover any variations, uses or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary techniques in the art that are not disclosed in this disclosure. The description and examples are to be considered exemplary only, and the true scope and spirit of the present invention are indicated by the following claims.

[0238] It should be understood that the present invention is not limited to the exact construction that has been described above and shown in the drawings and that various modifications and changes may be made without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

[0239] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

[0240] The above is a detailed introduction to a recommendation method, a recommendation device and a device for recommendation provided by the present invention. Specific examples are used in this article to illustrate the principle and implementation mode of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation mode and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A recommendation method, characterized in that: The method comprises: Obtain voice messages entered, sent or received by users through instant messaging applications; Performing voice recognition on the voice information to obtain text information corresponding to the voice information; Performing intent recognition on the text information to determine the user's recommendation intent; obtaining slot information corresponding to the recommendation intent; filling the slots of the text information according to the slot information to obtain each slot value in the text information corresponding to the slot information; Extracting features from the voice information to obtain voice features corresponding to the voice information; Determine, according to the voice feature, a personalized label of the voice information, wherein the personalized label includes at least any one of the following: emotion, health condition, and region; Generate a first vector corresponding to the recommendation intent according to the slot values ​​corresponding to the slot information in the text information, generate a second vector corresponding to the personalized tag, and a third vector corresponding to the voice feature; concatenate the first vector, the second vector, and the third vector to obtain a user feature vector; identify entity words in the text information, input the entity words as query words into a search engine, and retrieve web page information; generate a fourth vector according to the web page information corresponding to the text information; calculate the correlation between the user feature vector and the fourth vector, sort the web page information according to the correlation, and select a preset number of web page information with top sorting as recommended information; The recommendation information is displayed to the user in a chat interface of an instant messaging application currently used by the user.

2. The method according to claim 1, characterized in that The step of determining the personalized label of the voice information according to the voice feature includes: The speech feature is input into a classification model so as to output a personalized label corresponding to the speech feature through the classification model; wherein the classification model is trained based on speech sample data that has been labeled with personalized labels.

3. The method according to claim 2, characterized in that Before inputting the speech features into the classification model, the method further comprises: Acquire voice sample data, wherein the voice sample data is annotated with a corresponding personalized label; Extracting features from the speech sample data to obtain speech sample features corresponding to the speech sample data; Inputting the speech sample features into an initial classification model for training; During the training process, the model parameters of the initial classification model are adjusted until the personalized label output by the initial classification model matches the personalized label annotated with the speech sample data, thereby obtaining a trained classification model.

4. A speech recognition device, characterized in that: The device comprises: A voice acquisition module is used to acquire voice information input, sent or received by a user through an instant messaging application; A speech recognition module, used to perform speech recognition on the speech information to obtain text information corresponding to the speech information; An intention determination module, used to identify the intention of the text information and determine the recommendation intention of the user; a slot acquisition module, used to obtain the slot information corresponding to the recommendation intention; a slot filling module, used to fill the slots of the text information according to the slot information, and obtain the slot values ​​corresponding to the slot information in the text information; A feature extraction module, used to extract features from the voice information to obtain voice features corresponding to the voice information; A label determination module, used to determine a personalized label of the voice information according to the voice feature, wherein the personalized label includes at least one of the following: emotion, health condition, and region; An information recommendation module is used to generate a first vector corresponding to the recommendation intention, a second vector corresponding to the personalized tag, and a third vector corresponding to the voice feature according to the slot values ​​corresponding to the slot information in the text information; concatenate the first vector, the second vector, and the third vector to obtain a user feature vector; identify entity words in the text information, input the entity words as query words into a search engine, and retrieve web page information; generate a fourth vector according to the web page information corresponding to the text information; calculate the correlation between the user feature vector and the fourth vector, sort the web page information according to the correlation, and select a preset number of web page information with the highest sorting as recommended information; and display the recommended information to the user in a chat interface of an instant messaging application currently used by the user.

5. The device according to claim 4, characterized in that The label determination module is specifically used to input the speech feature into a classification model so as to output a personalized label corresponding to the speech feature through the classification model; wherein the classification model is trained based on speech sample data that has been labeled with personalized labels.

6. The device according to claim 5, characterized in that The device also includes: A sample acquisition module, used to acquire voice sample data, wherein the voice sample data is annotated with a corresponding personalized label; A feature acquisition module, used to extract features from the speech sample data to obtain speech sample features corresponding to the speech sample data; A feature training module, used for inputting the speech sample features into an initial classification model for training; The parameter adjustment module is used to adjust the model parameters of the initial classification model during the training process until the personalized label output by the initial classification model matches the personalized label annotated by the speech sample data, thereby obtaining a trained classification model.

7. A device for recommendation, characterized in that The device comprises a memory and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by one or more processors, wherein the one or more programs include instructions for performing the recommendation method as claimed in any one of claims 1 to 3.

8. A machine-readable medium having instructions stored thereon, which, when executed by one or more processors, causes a device to perform the recommendation method according to any one of claims 1 to 3.

Images