A standard problem recommendation method, device, medium and equipment

Abstract

The present invention relates to a method, device, medium and equipment for recommending standard questions. According to the solution provided by an embodiment of the present invention, based on the standard question sequence corresponding to each standard question that a specified user has asked in the past, a recommendation model obtained by fusing a recursive neural network model with a linear model that has been trained in advance can be used to determine the identifier of the standard question to be recommended for the specified user, and then recommend the corresponding standard question to the specified user. Therefore, the recommendation model obtained by fusing the recursive neural network model with the linear model can be used to quickly and accurately predict the standard question that the user may ask next, thereby achieving fast and accurate recommendation of standard questions.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a standard question recommendation method, device, medium and equipment. Background Art

[0002] This section is intended to provide a background or context to embodiments of the invention that are recited in the claims. No description herein is admitted to be prior art by inclusion in this section.

[0003] The intelligent question-answering system is a system that supports responses to various forms of questions through a standardized question-answering database. During the response process, the intelligent question-answering system needs to convert various forms of questions asked by users into standard questions, and then search for answers to standard questions in the question-answering database and respond.

[0004] In order to reduce the standard question conversion process and reduce the system load, the current intelligent question-answering system can recommend standard questions so that the user can directly find the answer from the question-answering library based on the standard question selected by the user.

[0005] However, current intelligent question-answering standard question recommendation schemes all have various problems, making it difficult to recommend standard questions quickly and accurately.

[0006] For example, content-based recommendation methods recommend similar questions based on questions that users have already asked, which makes it difficult to mine the user's next question intention, and standard question recommendations are not accurate enough. For another example, in collaborative filtering-based recommendation methods, the larger the number of users, the more complex the algorithm for calculating question similarity, which makes it difficult to ensure not only the timeliness of standard question recommendations, but also the accuracy of standard question recommendations. For another example, knowledge-based recommendation methods have a complex knowledge construction process and poor reusability, and it is also difficult to ensure the speed and accuracy of standard question recommendations.

[0007] Therefore, there is an urgent need to provide a solution that can quickly and accurately recommend standard questions. Summary of the invention

[0008] The embodiments of the present invention provide a standard question recommendation method, apparatus, medium and device, which are used to solve the problem that standard questions cannot be recommended quickly and accurately.

[0009] In a first aspect, the present invention provides a standard question recommendation method, the method comprising:

[0010] Determine a standard question sequence corresponding to a specified user identifier, where the standard question sequence is obtained by arranging the identifiers of each standard question in the history of each user asking the specified user identifier in sequence according to the time of asking the question;

[0011] Determining a first user feature vector corresponding to the standard question sequence;

[0012] Taking the first user feature vector corresponding to the determined standard question sequence as input, and determining the identifier of the standard question to be recommended through a pre-trained recommendation model;

[0013] Recommending the standard question corresponding to the determined identifier of the standard question to be recommended to the user corresponding to the designated user identifier;

[0014] The recommendation model is a model obtained by fusing a recursive neural network model with a linear model.

[0015] Optionally, the method further includes:

[0016] Determine user attribute information corresponding to the specified user identifier;

[0017] Determine a second user feature vector corresponding to the user attribute information;

[0018] The first user feature vector corresponding to the determined standard question sequence is used as input, and the identifier of the standard question to be recommended is determined by a pre-trained recommendation model, including:

[0019] The first user feature vector corresponding to the determined standard question sequence and the second user feature vector corresponding to the user attribute information are taken as input, and the identifier of the standard question to be recommended is determined through a pre-trained recommendation model.

[0020] Optionally, a pre-trained recommendation model that identifies standard questions to recommend, including:

[0021] Using a linear model to obtain a linear vector according to the first user feature vector, and using a hidden layer of a recursive neural network model to obtain a recursive vector according to the first user feature vector;

[0022] Merging the recursive vector with a second user feature vector corresponding to the user attribute information to obtain a merged vector;

[0023] A logistic regression calculation is performed on the linear vector and the merged vector according to the logistic loss function corresponding to the recursive neural network model to determine the identifier of the standard question to be recommended.

[0024] Optionally, the recursive neural network model is a long short-term memory network model.

[0025] Optionally, obtaining a linear vector according to the first user feature vector by using a linear model includes:

[0026] A linear vector is obtained according to the first user feature vector and / or the cross feature vector using a linear model, wherein each cross feature corresponding to the cross feature vector is obtained by crossing the first user feature corresponding to the first user feature vector.

[0027] Optionally, the first user feature X corresponding to the first user feature vector is cross-linked to obtain the kth cross-feature φ corresponding to the cross-feature vector in the following manner: k (X):

[0028]

[0029] Among them, X i ~X d d first user features corresponding to the selected first user feature vector;

[0030] C ki Indicates whether the i-th first user feature among the selected d first user features participates in the construction of the k-th cross user feature.

[0031] In a second aspect, the present invention further provides a standard question recommendation device, the device comprising:

[0032] A collection module, used to determine a standard question sequence corresponding to a specified user identifier, wherein the standard question sequence is obtained by arranging the identifiers of each standard question in the order of the time of asking the question in the history of each user corresponding to the specified user identifier;

[0033] An analysis module, used to determine a first user feature vector corresponding to the standard question sequence;

[0034] A recommendation module is used to take the first user feature vector corresponding to the determined standard question sequence as input, determine the identifier of the standard question to be recommended through a pre-trained recommendation model; and recommend the standard question corresponding to the determined identifier of the standard question to be recommended to the user corresponding to the specified user identifier; wherein the recommendation model is a model obtained by fusing a recursive neural network model with a linear model.

[0035] Optionally, the acquisition module is further used to determine user attribute information corresponding to the designated user identifier;

[0036] The analysis module is further used to determine a second user feature vector corresponding to the user attribute information;

[0037] The recommendation module uses the first user feature vector corresponding to the determined standard question sequence as input, and determines the identifier of the standard question to be recommended through a pre-trained recommendation model, including:

[0038] The first user feature vector corresponding to the determined standard question sequence and the second user feature vector corresponding to the user attribute information are taken as input, and the identifier of the standard question to be recommended is determined through a pre-trained recommendation model.

[0039] Optionally, a pre-trained recommendation model that identifies standard questions to recommend, including:

[0040] Using a linear model to obtain a linear vector according to the first user feature vector, and using a hidden layer of a recursive neural network model to obtain a recursive vector according to the first user feature vector;

[0041] Merging the recursive vector with a second user feature vector corresponding to the user attribute information to obtain a merged vector;

[0042] A logistic regression calculation is performed on the linear vector and the merged vector according to the logistic loss function corresponding to the recursive neural network model to determine the identifier of the standard question to be recommended.

[0043] Optionally, the recursive neural network model is a long short-term memory network model.

[0044] Optionally, obtaining a linear vector according to the first user feature vector by using a linear model includes:

[0045] A linear vector is obtained according to the first user feature vector and / or the cross feature vector using a linear model, wherein each cross feature corresponding to the cross feature vector is obtained by crossing the first user feature corresponding to the first user feature vector.

[0046] Optionally, the first user feature X corresponding to the first user feature vector is cross-linked to obtain the kth cross-feature φ corresponding to the cross-feature vector in the following manner: k (X):

[0047]

[0048] Among them, X i ~X d d first user features corresponding to the selected first user feature vector;

[0049] C ki Indicates whether the i-th first user feature among the selected d first user features participates in the construction of the k-th cross user feature.

[0050] In a third aspect, the present invention further provides a non-volatile computer storage medium, wherein the computer storage medium stores an executable program, and the executable program is executed by a processor to implement the method as described above.

[0051] In a fourth aspect, the present invention further provides a standard question recommendation device, comprising 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;

[0052] The memory is used to store computer programs;

[0053] The processor is used to implement the method steps described above when executing the program stored in the memory.

[0054] According to the solution provided by the embodiment of the present invention, based on the standard question sequence corresponding to each standard question that the designated user has asked in the past, the identification of the standard question to be recommended for the designated user can be determined by using the recommendation model obtained by fusing the recursive neural network model with the linear model, and then the corresponding standard question can be recommended to the designated user. Therefore, the recommendation model obtained by fusing the recursive neural network model with the linear model can be used to quickly and accurately predict the standard question that the user may ask next, thereby achieving fast and accurate recommendation of standard questions.

[0055] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description, claims, and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] 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 will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0057] Figure 1 A schematic diagram of a flow chart of a standard question recommendation method provided by an embodiment of the present invention;

[0058] Figure 2 A schematic diagram of the structure of a pre-trained recommendation model provided in an embodiment of the present invention;

[0059] Figure 3 A schematic diagram of the structure of a pre-trained recommendation model provided in an embodiment of the present invention;

[0060] Figure 4 A schematic diagram of the structure of an improved long short-term memory network model provided by an embodiment of the present invention;

[0061] Figure 5 A schematic diagram of the structure of a standard question recommendation device provided by an embodiment of the present invention;

[0062] Figure 6 A schematic diagram of the relationship between a user, a standard question recommendation device, an intelligent question answering system, and a manual question answering system provided by an embodiment of the present invention;

[0063] Figure 7 A schematic diagram of the structure of a standard question recommendation device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0064] In order to make the purpose, technical scheme and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all 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.

[0065] It should be noted that the "multiple or several" mentioned in this article refers to two or more. "And / or" describes the association relationship of the associated objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the associated objects before and after are in an "or" relationship.

[0066] The terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in sequences other than those illustrated or described herein.

[0067] In addition, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus that includes a series of steps or elements is not necessarily limited to those steps or elements explicitly listed, but may include other steps or elements not explicitly listed or inherent to such process, method, product, or apparatus.

[0068] Taking into account that the recursive neural network model can realize deep memory function and the linear model can realize shallow memory function, this application considers obtaining a recommendation model by fusing the recursive neural network model with the linear model, and predicting and recommending the standard questions that the user may ask next according to the first user feature vector corresponding to the standard question sequence corresponding to the user, so as to realize fast and accurate recommendation of standard questions in the intelligent question and answer system.

[0069] In the present application, the fused linear model in the recommendation model can not only obtain a vector representation based on the first user feature vector, but can also further obtain a cross feature based on the first user feature in the first user feature vector, and then obtain a vector representation based on the cross feature vector (or based on the cross feature vector and the first user feature vector). Thus, the interaction between the original features can be captured through the cross feature, and the function of nonlinear memory is added on the basis that the linear model can realize linear memory, so as to better realize the shallow memory function.

[0070] In addition, in the present application, the recursive neural network model integrated in the recommendation model can be a long short-term memory network model, so as to realize the memory function of short-term historical information and long-term historical information, and better realize the deep memory function.

[0071] In addition, in the present application scheme, the vector representation for performing logistic regression operations based on the recursive neural network model in the recommendation model can be obtained directly based on the recursive neural network model, and can also be further combined with the vector representation corresponding to the user attribute information to obtain a richer vector representation, so that the recommendation of standard questions is faster and more accurate.

[0072] Based on the above description, an embodiment of the present invention provides a standard question recommendation method, the steps of which can be as follows: Figure 1 As shown, including:

[0073] Step 101: Determine a standard question sequence corresponding to a specified user identifier.

[0074] In this step, a standard question sequence corresponding to the specified user identifier can be determined, and the standard question sequence is obtained by arranging the identifiers of each standard question in the order of the time of asking each standard question in the history of the user corresponding to the specified user identifier. The identifier of each standard question can be used to uniquely identify a standard question, so that the purpose of simplifying the standard question sequence can be achieved by representing each standard question with a corresponding identifier.

[0075] In a possible implementation, each standard question historically asked by a user corresponding to a specified user ID may be obtained from log information. The log information may be stored in a relational database.

[0076] Obtaining standard questions from log information can be achieved through relevant processing of log information, such as data collection, cleaning, standardization, feature combination and extraction.

[0077] That is to say, in this step, for the user who needs standard questions recommended, each standard question asked by the user in the past can be obtained according to the user's designated user ID, and each standard question can be arranged in sequence according to the time of questioning to obtain the standard question sequence corresponding to the user.

[0078] Taking the user as an enterprise user as an example, the user's designated user identifier may be, but is not limited to, the user name corresponding to the enterprise user and the intelligent question-answering system. For example, taking the intelligent question-answering system set up on a certain website as an example, the user's designated user identifier may be the user name registered by the enterprise user on the website; taking the intelligent question-answering system set up on a certain client as an example, the user's designated user identifier may be the user name registered by the enterprise user for the client.

[0079] Step 102: Determine a first user feature vector corresponding to the standard question sequence.

[0080] In this step, the obtained standard question sequence can be converted into a standardized feature vector that meets the requirements of mathematical modeling.

[0081] Step 103: Taking the first user feature vector corresponding to the determined standard question sequence as input, the identifier of the standard question to be recommended is determined through the pre-trained recommendation model.

[0082] In this step, the identification of the standard question to be recommended can be determined by a pre-trained recommendation model according to the first user feature vector corresponding to the standard question sequence.

[0083] In this embodiment, the recommendation model is a model obtained by fusing a recursive neural network model with a linear model.

[0084] In a possible implementation, the recursive neural network model integrated in the recommendation model may adopt a long short-term memory network model.

[0085] In a possible implementation, the structural diagram of the pre-trained recommendation model can be as follows: Figure 2 As shown. Figure 2 As shown, the recursive neural network model in the recommendation model is a long short-term memory network model.

[0086] like Figure 2 As shown, the recursive neural network model in the recommendation model can adopt a three-layer long short-term memory network model, which includes a hidden layer consisting of a first long short-term memory network layer, a second long short-term memory network layer and a third long short-term memory network layer.

[0087] It can be understood that the standard question sequence corresponding to user question 1 to user question n corresponds to a high-dimensional sparse matrix, and a recursive vector can be obtained according to the first user feature vector corresponding to the standard question sequence by using the hidden layer of the recursive neural network model.

[0088] Among them, after one-hot encoding is performed on each user question, a dense vector representation is performed through embedding vectorization, and the obtained vector can be used as the input of the hidden layer to obtain a recursive vector.

[0089] In addition, a linear model may be used to obtain a linear vector according to the first user feature vector corresponding to the standard question sequence.

[0090] Then, the linear vector and the recursive vector can be subjected to logistic regression calculation according to the logistic loss function (softmax layer) corresponding to the recursive neural network model, and the identification of the standard problem to be recommended can be determined and output.

[0091] Combination Figure 2 Given a schematic diagram of the recommendation model structure, in one possible implementation, the pre-trained recommendation model determines the identification of the standard question to be recommended, which may include:

[0092] Using a linear model to obtain a linear vector according to the first user feature vector, and using a hidden layer of a recursive neural network model to obtain a recursive vector according to the first user feature vector;

[0093] Logistic regression calculation is performed on the linear vector and the recursive vector according to the logistic loss function corresponding to the recursive neural network model to determine the identification of the standard problem to be recommended.

[0094] The training process of the recommendation model is similar to the use process of the recommendation model, wherein the recommendation model can be trained in combination with the manually annotated standard questions to be recommended (identification), which will not be further described in this embodiment. After continuously acquiring training samples to train the recommendation model, the trained recommendation model can obtain a deep question-answer relationship, and can obtain the identification of the user's potential standard questions that may be asked next based on the standard question sequence corresponding to the user, and then can recommend the corresponding standard questions to the user for selection based on the obtained identification.

[0095] Step 104: recommend the standard question corresponding to the determined identifier of the standard question to be recommended to the user corresponding to the designated user identifier.

[0096] In this step, the corresponding standard questions can be determined according to the determined identifier, and then the determined standard questions can be recommended to users who need standard question recommendations.

[0097] It should be noted that in step 103, the number of determined identifiers may be at least one. In this step, the standard questions corresponding to each determined identifier may be recommended to the user, or the standard questions corresponding to some of the multiple determined identifiers may be recommended to the user.

[0098] In a possible implementation, before step 103, user attribute information corresponding to the designated user identifier may be determined, and a second user feature vector corresponding to the user attribute information may be determined (i.e., the obtained user attribute information is converted into a standard feature vector that meets the requirements of mathematical modeling). That is, the attribute information of the user for whom standard questions are to be recommended may be determined, and the corresponding feature vector may be determined, and then the standard questions may be recommended in combination with the user attribute information, thereby further improving the accuracy of the recommendation.

[0099] Taking the user as an enterprise user as an example, the user's attribute information may include but is not limited to at least one of the enterprise tax number, enterprise address, enterprise type, enterprise industry, enterprise scale, enterprise location, enterprise registered capital, and device version used by the enterprise.

[0100] In a possible implementation, the user's attribute information can also be obtained through log information, and the log information can be stored in a relational database. The user's attribute information can also be obtained from the log information by related processing of the log information, such as data collection, cleaning, standardization, feature combination and extraction.

[0101] At this time, in step 103, the first user feature vector corresponding to the determined standard question sequence and the second user feature vector corresponding to the user attribute information can be used as input, and the identifier of the standard question to be recommended can be determined through the pre-trained recommendation model.

[0102] That is to say, at this time in step 103, the identification of the standard question to be recommended can be determined by the pre-trained recommendation model according to the first user feature vector corresponding to the standard question sequence and the second user feature vector corresponding to the user attribute information.

[0103] In a possible implementation, the structural diagram of the pre-trained recommendation model can be as follows: Figure 3 As shown. Figure 3 As shown in , the recursive neural network model in the recommendation model is a long short-term memory network model, such as Figure 3 The recommendation model shown can be understood as a model obtained by integrating the long short-term memory network model obtained by improving the long short-term memory network model in combination with user attribute information and the linear model.

[0104] like Figure 3As shown, the recursive neural network model in the recommendation model can adopt a three-layer long short-term memory network model, which includes a hidden layer consisting of a first long short-term memory network layer, a second long short-term memory network layer and a third long short-term memory network layer.

[0105] It can be understood that the standard question sequence corresponding to user question 1 to user question n corresponds to a high-dimensional sparse matrix, and a recursive vector can be obtained according to the first user feature vector corresponding to the standard question sequence using the hidden layer of the recursive neural network model.

[0106] Among them, after one-hot encoding is performed on each user question, a dense vector representation is performed through embedding vectorization, and the obtained vector can be used as the input of the hidden layer to obtain a recursive vector.

[0107] Assuming that the user who needs to be recommended for standard questions is an enterprise user, the user attribute information obtained includes the enterprise type, the industry to which the enterprise belongs, the enterprise scale, the region where the enterprise is located, the enterprise registered capital, and the device version used by the enterprise. The user attribute information can be understood as corresponding to a low-dimensional sparse matrix. Figure 3 As shown, the recursive vector and the second user feature vector corresponding to the user attribute information may be merged to obtain a merged vector.

[0108] In addition, a linear model may be used to obtain a linear vector according to the first user feature vector corresponding to the standard question sequence.

[0109] Then, the linear vector and the merged vector can be subjected to logistic regression calculation according to the logistic loss function (softmax layer) corresponding to the recursive neural network model to determine the identification of the standard problem to be recommended and output it.

[0110] Figure 3 In the diagram, the structural diagram of the long short-term memory network model obtained by improving the long short-term memory network model through user attribute information can be shown as follows Figure 4 shown.

[0111] Combination Figure 3 Given a schematic diagram of the recommendation model structure, in one possible implementation, the pre-trained recommendation model determines the identification of the standard question to be recommended, which may include:

[0112] Using a linear model to obtain a linear vector according to the first user feature vector, and using a hidden layer of a recursive neural network model to obtain a recursive vector according to the first user feature vector;

[0113] Merging the recursive vector with the second user feature vector corresponding to the user attribute information to obtain a merged vector;

[0114] Logistic regression calculation is performed on the linear vector and the merged vector according to the logistic loss function corresponding to the recursive neural network model to determine the identification of the standard question to be recommended.

[0115] In a possible implementation manner, obtaining a linear vector according to the first user feature vector by using a linear model includes:

[0116] A linear vector is obtained according to the first user feature vector and / or the cross feature vector using a linear model, and each cross feature corresponding to the cross feature vector is obtained by crossing the first user feature corresponding to the first user feature vector.

[0117] Further, in a possible implementation, the first user feature X corresponding to the first user feature vector may be cross-linked to obtain the kth cross-feature φ corresponding to the cross-feature vector in the following manner: k (X):

[0118]

[0119] Among them, X i ~X d d first user features corresponding to the selected first user feature vector;

[0120] C ki Indicates whether the i-th first user feature among the selected d first user features participates in the construction of the k-th cross user feature.

[0121] In the solution provided by the embodiment of the present invention, a method of fusing shallow features with deep features is creatively proposed, and a recommendation model obtained by fusing a recursive neural network model with a linear model is used to recommend standard questions. In addition, by combining user attribute information with user historical question records, the standard questions that the user wants to ask next can be inferred based on the user attribute information and the user historical question records, and standard questions can be recommended in advance, which ensures the timeliness and accuracy of standard question recommendations while solving the cold start problem existing in traditional neural network recommendation methods.

[0122] For example, a standard question that a user has asked in the past is "How do I issue an invoice?", and the answer to this standard question is "Issuing an invoice through the Golden Tax Disk". According to the solution provided by an embodiment of the present invention, the standard questions recommended to the user through the pre-trained recommendation model may be related standard questions such as "How do I get a Golden Tax Disk?", "What is a Golden Tax Disk?", and "How do I go to the tax bureau to apply for a Golden Tax Disk?", rather than similar standard questions such as "How do I issue an invoice?"

[0123] In the solution provided by the embodiment of the present invention, the user can be an enterprise user, that is, starting from the intelligent question-answering business scenario of enterprise users, the user attribute information and historical standard question records of enterprise users can be extracted, and shallow cross-features can be constructed. By combining shallow cross-features with long-term and short-term memory networks, deep information in the historical standard question records of enterprise users can be obtained while hot starting, and pre-recommendation can be achieved. It can be better applied to the early and late stages of the entire intelligent question-answering system, and it is no longer a simple similar recommendation of questions that users have asked, but an intelligent pre-recommendation based on deep historical information.

[0124] Corresponding to the provided method, the following device is further provided.

[0125] The embodiment of the present invention provides a standard question recommendation device, the structure of which can be as follows: Figure 5 As shown, including:

[0126] The acquisition module 11 is used to determine a standard question sequence corresponding to a specified user identifier, where the standard question sequence is obtained by arranging the identifiers of each standard question in the order of the time of asking the question in the history of each user corresponding to the specified user identifier;

[0127] The analysis module 12 is used to determine the first user feature vector corresponding to the standard question sequence;

[0128] The recommendation module 13 is used to take the first user feature vector corresponding to the determined standard question sequence as input, and determine the identifier of the standard question to be recommended through a pre-trained recommendation model; and recommend the standard question corresponding to the determined identifier of the standard question to be recommended to the user corresponding to the specified user identifier; wherein the recommendation model is a model obtained by fusing a recursive neural network model with a linear model.

[0129] Optionally, the acquisition module 11 is further used to determine user attribute information corresponding to the designated user identifier;

[0130] The analysis module 12 is also used to determine a second user feature vector corresponding to the user attribute information;

[0131] The recommendation module 13 uses the first user feature vector corresponding to the determined standard question sequence as input, and determines the identifier of the standard question to be recommended through a pre-trained recommendation model, including:

[0132] The first user feature vector corresponding to the determined standard question sequence and the second user feature vector corresponding to the user attribute information are taken as input, and the identifier of the standard question to be recommended is determined through a pre-trained recommendation model.

[0133] Optionally, a pre-trained recommendation model that identifies standard questions to recommend, including:

[0134] Using a linear model to obtain a linear vector according to the first user feature vector, and using a hidden layer of a recursive neural network model to obtain a recursive vector according to the first user feature vector;

[0135] Merging the recursive vector with a second user feature vector corresponding to the user attribute information to obtain a merged vector;

[0136] A logistic regression calculation is performed on the linear vector and the merged vector according to the logistic loss function corresponding to the recursive neural network model to determine the identifier of the standard question to be recommended.

[0137] Optionally, the recursive neural network model is a long short-term memory network model.

[0138] Optionally, obtaining a linear vector according to the first user feature vector by using a linear model includes:

[0139] A linear vector is obtained according to the first user feature vector and / or the cross feature vector using a linear model, wherein each cross feature corresponding to the cross feature vector is obtained by crossing the first user feature corresponding to the first user feature vector.

[0140] Optionally, the first user feature X corresponding to the first user feature vector is cross-linked to obtain the kth cross-feature φ corresponding to the cross-feature vector in the following manner: k (X):

[0141]

[0142] Among them, X i ~X d d first user features corresponding to the selected first user feature vector;

[0143] C ki Indicates whether the i-th first user feature among the selected d first user features participates in the construction of the k-th cross user feature.

[0144] Usually, the manual question answering system and the intelligent question answering system can complement each other to solve problems for users. The relationship diagram between the user, the standard question recommendation device, the intelligent question answering system and the manual question answering system can be, but is not limited to, as follows: Figure 6 shown.

[0145] The collection module of the standard question recommendation device can determine the standard question sequence and user attribute information corresponding to the specified user identifier from the intelligent question and answer system, for example, the log information corresponding to the intelligent question and answer system. After the recommendation module obtains the feature vector sent by the analysis module, the determined standard questions can be recommended to users, such as corporate users, and the determined standard questions can be sent to the manual question and answer system to assist manual customer service in referring to the standard questions that users may need to ask.

[0146] The functions of each functional unit of each device provided in the above embodiments of the present invention can be implemented through the steps of the above corresponding methods. Therefore, the specific working process and beneficial effects of each functional unit in each device provided in the embodiments of the present invention are not repeated here.

[0147] Based on the same inventive concept, embodiments of the present invention provide the following devices and media.

[0148] The embodiment of the present invention provides a standard question recommendation device, the structure of which can be as follows: Figure 7 As shown, it includes a processor 21, a communication interface 22, a memory 23 and a communication bus 24, wherein the processor 21, the communication interface 22, and the memory 23 communicate with each other through the communication bus 24;

[0149] The memory 23 is used to store computer programs;

[0150] The processor 21 is used to implement the steps described in the above method embodiment of the present invention when executing the program stored in the memory.

[0151] Optionally, the processor 21 may specifically include a central processing unit (CPU), an application specific integrated circuit (ASIC), may be one or more integrated circuits for controlling program execution, may be a hardware circuit developed using a field programmable gate array (FPGA), or may be a baseband processor.

[0152] Optionally, the processor 21 may include at least one processing core.

[0153] Optionally, the memory 23 may include a read-only memory (ROM), a random access memory (RAM) and a disk memory. The memory 23 is used to store data required by at least one processor 21 when running. The number of memories 23 may be one or more.

[0154] An embodiment of the present invention further provides a non-volatile computer storage medium, wherein the computer storage medium stores an executable program. When the executable program is executed by a processor, the method provided by the above method embodiment of the present invention is implemented.

[0155] In the specific implementation process, the computer storage medium may include: Universal Serial Bus Flash Drive (USB), mobile hard disk, read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk, and other storage media that can store program codes.

[0156] In the embodiments of the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the units or division of units are only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection of devices or units can be electrical or other forms.

[0157] The functional units in the embodiment of the present invention may be integrated into one processing unit, or the units may be independent physical modules.

[0158] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the technical solution of the embodiment of the present invention can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device, such as a personal computer, a server, or a network device, or a processor to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: a Universal Serial Bus Flash Drive, a mobile hard disk, a ROM, a RAM, a magnetic disk or an optical disk, and other media that can store program codes.

[0159] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0160] The present invention is described with reference to flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0161] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0162] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0163] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0164] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A standard problem recommendation method, It is characterized in that The method comprises: Determine a standard question sequence corresponding to a specified user identifier, where the standard question sequence is obtained by arranging the identifiers of each standard question in the history of each user asking the specified user identifier in sequence according to the time of asking the question; Determining a first user feature vector corresponding to the standard question sequence; Taking the first user feature vector corresponding to the determined standard question sequence as input, and determining the identifier of the standard question to be recommended through a pre-trained recommendation model; Recommending the standard question corresponding to the determined identifier of the standard question to be recommended to the user corresponding to the designated user identifier; The recommendation model is a model obtained by fusing a recursive neural network model with a linear model.

2. The method according to claim 1, It is characterized in that The method further comprises: Determine user attribute information corresponding to the specified user identifier; Determine a second user feature vector corresponding to the user attribute information; The first user feature vector corresponding to the determined standard question sequence is used as input, and the identifier of the standard question to be recommended is determined by a pre-trained recommendation model, including: The first user feature vector corresponding to the determined standard question sequence and the second user feature vector corresponding to the user attribute information are taken as input, and the identifier of the standard question to be recommended is determined through a pre-trained recommendation model.

3. The method according to claim 2, It is characterized in that The pre-trained recommendation model identifies the standard questions to be recommended, including: Using a linear model to obtain a linear vector according to the first user feature vector, and using a hidden layer of a recursive neural network model to obtain a recursive vector according to the first user feature vector; Merging the recursive vector with a second user feature vector corresponding to the user attribute information to obtain a merged vector; A logistic regression calculation is performed on the linear vector and the merged vector according to the logistic loss function corresponding to the recursive neural network model to determine the identifier of the standard question to be recommended.

4. The method according to claim 3, It is characterized in that The recursive neural network model is a long short-term memory network model.

5. The method according to claim 3, It is characterized in that Obtaining a linear vector according to the first user feature vector using a linear model includes: A linear vector is obtained according to the first user feature vector and / or the cross feature vector using a linear model, wherein each cross feature corresponding to the cross feature vector is obtained by crossing the first user feature corresponding to the first user feature vector.

6. The method according to claim 5, It is characterized in that The first user feature X corresponding to the first user feature vector is crossed to obtain the kth cross feature φ corresponding to the cross feature vector. k (X): Among them, X i ~X d d first user features corresponding to the selected first user feature vector; C ki Indicates whether the i-th first user feature among the selected d first user features participates in the construction of the k-th cross user feature.

7. A standard question recommendation device, It is characterized in that The device comprises: A collection module, used to determine a standard question sequence corresponding to a specified user identifier, wherein the standard question sequence is obtained by arranging the identifiers of each standard question in the order of the time of asking the question in the history of each user corresponding to the specified user identifier; An analysis module, used to determine a first user feature vector corresponding to the standard question sequence; A recommendation module is used to take the first user feature vector corresponding to the determined standard question sequence as input, determine the identifier of the standard question to be recommended through a pre-trained recommendation model; and recommend the standard question corresponding to the determined identifier of the standard question to be recommended to the user corresponding to the specified user identifier; wherein the recommendation model is a model obtained by fusing a recursive neural network model with a linear model.

8. The device according to claim 7, It is characterized in that The acquisition module is further used to determine the user attribute information corresponding to the designated user identifier; The analysis module is further used to determine a second user feature vector corresponding to the user attribute information; The recommendation module uses the first user feature vector corresponding to the determined standard question sequence as input, and determines the identifier of the standard question to be recommended through a pre-trained recommendation model, including: The first user feature vector corresponding to the determined standard question sequence and the second user feature vector corresponding to the user attribute information are taken as input, and the identifier of the standard question to be recommended is determined through a pre-trained recommendation model.

9. The device as claimed in claim 8, It is characterized in that The pre-trained recommendation model identifies the standard questions to be recommended, including: Using a linear model to obtain a linear vector according to the first user feature vector, and using a hidden layer of a recursive neural network model to obtain a recursive vector according to the first user feature vector; Merging the recursive vector with a second user feature vector corresponding to the user attribute information to obtain a merged vector; A logistic regression calculation is performed on the linear vector and the merged vector according to the logistic loss function corresponding to the recursive neural network model to determine the identifier of the standard question to be recommended.

10. The device according to claim 9, It is characterized in that The recursive neural network model is a long short-term memory network model.

11. The device according to claim 9, It is characterized in that Obtaining a linear vector according to the first user feature vector using a linear model includes: A linear vector is obtained according to the first user feature vector and / or the cross feature vector using a linear model, wherein each cross feature corresponding to the cross feature vector is obtained by crossing the first user feature corresponding to the first user feature vector.

12. The device according to claim 11, It is characterized in that The first user feature X corresponding to the first user feature vector is crossed to obtain the kth cross feature φ corresponding to the cross feature vector. k (X): Among them, X i ~X d d first user features corresponding to the selected first user feature vector; C ki Indicates whether the i-th first user feature among the selected d first user features participates in the construction of the k-th cross user feature.

13. A non-volatile computer storage medium, It is characterized in that The computer storage medium stores an executable program, and the executable program is executed by a processor to implement any one of the methods of claims 1 to 6.

14. A standard issue recommended device, It is characterized in that The device comprises 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; The memory is used to store computer programs; The processor is used to implement the method steps described in any one of claims 1 to 6 when executing the program stored in the memory.

Images