Model training and product information query method and device, equipment, medium and product

Abstract

The invention discloses a model training and product information query method and device, equipment, a medium and a product, and relates to the technical field of artificial intelligence, and the method comprises the steps: determining at least one-order neighbor node corresponding to a target node as a target neighbor node according to a product relation network; according to the feature vectors corresponding to the target neighbor nodes, updating the feature vectors corresponding to the target nodes, and generating updated feature vectors corresponding to the target nodes; using the product structured information corresponding to the target node as a sample query word, using the updated feature vector as a sample feature vector corresponding to the sample query word, and training the to-be-trained model according to the sample query word and the sample feature vector to generate a feature vector prediction model; wherein the feature vector prediction model is used for predicting a corresponding target feature vector according to an original query word of a user, and the target feature vector is used for querying product information from a product specification. According to the invention, the effect of improving the accuracy of product information query is realized.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a model training and product information query method, device, equipment, medium and product. Background Art

[0002] With the development of artificial intelligence technology, more and more business scenarios that originally required human participation can now be realized using artificial intelligence technology, which undoubtedly greatly saves labor costs and improves the efficiency of business execution. Among them, the product information query scenario is a typical business scenario that can be realized using artificial intelligence technology.

[0003] In current product information query scenarios, usually only the query words given by the user are used to query information in the knowledge base. However, when the quality of the query words given by the user is poor, only using the query words to query information will result in inaccurate recalled product information, affecting the user experience. Summary of the invention

[0004] The present invention provides a model training and product information query method, device, equipment, medium and product to solve the problem that the recalled product information is not accurate enough when querying the current product information.

[0005] According to one aspect of the present invention, a model training method is provided, comprising:

[0006] Determining at least one type of product structured information included in at least one product specification, and constructing a product relationship network based on the product structured information;

[0007] Determine any node from the product relationship network as a target node, and determine at least one first-order neighbor node corresponding to the target node as a target neighbor node according to the product relationship network;

[0008] According to the feature vectors corresponding to each of the target neighbor nodes, the feature vector corresponding to the target node is updated to generate an updated feature vector corresponding to the target node;

[0009] Taking the product structured information corresponding to the target node as a sample query word, and taking the updated feature vector as a sample feature vector corresponding to the sample query word, and training a to-be-trained model according to the sample query word and the sample feature vector to generate a feature vector prediction model;

[0010] The feature vector prediction model is used to predict the corresponding target feature vector according to the user's original query word, and the target feature vector is used to query product information from the product manual.

[0011] According to another aspect of the present invention, there is provided a product information query method, comprising:

[0012] Determine an original query word according to the user's query sentence, and input the original query word into a feature vector prediction model to predict a target feature vector corresponding to the original query word; wherein the feature vector prediction model is trained using the model training method as described above;

[0013] A target standard query word is determined from candidate standard query words according to the target feature vector, and the target standard query word is used to query product information from at least one product specification; wherein the candidate standard query word is determined according to product structured information included in the product specification.

[0014] According to another aspect of the present invention, there is provided a model training device, comprising:

[0015] A product relationship network construction module, used to determine at least one type of product structured information included in at least one product specification, and to construct a product relationship network according to the product structured information;

[0016] A target neighbor node determination module, used to determine any node from the product relationship network as a target node, and determine at least one first-order neighbor node corresponding to the target node as a target neighbor node according to the product relationship network;

[0017] An updated feature vector generation module is used to update the feature vector corresponding to the target node according to the feature vectors corresponding to each of the target neighbor nodes, and generate an updated feature vector corresponding to the target node;

[0018] A model training module, used to use the product structured information corresponding to the target node as a sample query word, and the updated feature vector as a sample feature vector corresponding to the sample query word, and train a to-be-trained model according to the sample query word and the sample feature vector to generate a feature vector prediction model;

[0019] The feature vector prediction model is used to predict the corresponding target feature vector according to the user's original query word, and the target feature vector is used to query product information from the product manual.

[0020] According to another aspect of the present invention, there is provided a product information query device, comprising:

[0021] A feature vector prediction module is used to determine the original query word according to the user's query sentence, and input the original query word into the feature vector prediction model to predict the target feature vector corresponding to the original query word; wherein the feature vector prediction model is trained using the model training method as described above;

[0022] A product information query module is used to determine a target standard query word from candidate standard query words according to the target feature vector, and use the target standard query word to query product information from at least one product specification; wherein the candidate standard query word is determined according to the product structured information included in the product specification.

[0023] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0024] at least one processor; and

[0025] a memory communicatively connected to the at least one processor; wherein,

[0026] The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the model training method and / or product information query method described in any one of the present invention.

[0027] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the model training method and / or product information query method described in any one of the present invention when executed.

[0028] According to another aspect of the present invention, a computer program product is provided, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements the model training method and / or product information query method described in any one of the present invention.

[0029] The present invention determines at least one type of product structured information included in at least one product description, and constructs a product relationship network according to the product structured information; determines any node from the product relationship network as a target node, and determines at least one first-order neighbor node corresponding to the target node as a target neighbor node according to the product relationship network; updates the feature vector corresponding to the target node according to the feature vectors corresponding to each target neighbor node, and generates an updated feature vector corresponding to the target node; uses the product structured information corresponding to the target node as a sample query word, and uses the updated feature vector as a sample feature vector corresponding to the sample query word, and trains a to-be-trained model according to the sample query word and the sample feature vector, and generates a feature vector prediction model. type; wherein, the feature vector prediction model is used to predict the corresponding target feature vector according to the user's original query term, and the target feature vector is used to query product information from the product manual. Since the updated feature vector corresponding to the target node not only contains its own feature information, but also includes the feature information of its corresponding target neighbor node, the sample feature vector determined according to the updated feature vector contains richer feature information, and further enables the feature vector prediction model trained according to the sample feature vector to predict the target feature vector containing rich feature information according to the original query term. Since the target feature vector has rich feature information, it is ultimately guaranteed that more accurate product information can be queried according to the target feature vector, thereby ensuring user experience.

[0030] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. 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 creative work.

[0032] Figure 1 A flowchart of a model training method provided in Embodiment 1 of the present invention;

[0033] Figure 2 A flowchart of a model training method provided in Embodiment 2 of the present invention;

[0034] Figure 3 A flowchart of a product information query method provided in Embodiment 3 of the present invention;

[0035] Figure 4A schematic diagram of the structure of a model training device provided in Embodiment 4 of the present invention;

[0036] Figure 5 A schematic diagram of the structure of a product information query device provided in Embodiment 5 of the present invention;

[0037] Figure 6 A schematic diagram of the structure of an electronic device that can be used to implement an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0038] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0039] It should be noted that the terms "target", "auxiliary", "sample", "original", "first", "second", "third", "fourth", 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 data used in this way can be interchangeable where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0040] Embodiment 1

[0041] Figure 1 This is a flow chart of a model training method provided in the first embodiment of the present invention. This embodiment can be applied to training a feature vector prediction model for predicting the target feature vector corresponding to the user's original query word, and the target feature vector is used to query product information from the product manual. The method can be executed by a model training device, which can be implemented in the form of hardware and / or software, such as a server. Figure 1 As shown, the method includes:

[0042] S101. Determine at least one type of product structured information included in at least one product specification, and construct a product relationship network based on the product structured information.

[0043] Among them, the product manual refers to a file that records the product information corresponding to a certain product. For example, the type of the product manual can be a drug manual. In this case, the product manual is a file that records the drug information corresponding to a certain drug. This embodiment does not limit the specific type of the product manual.

[0044] Product structured information refers to structured information obtained by extracting structured information from the content recorded in the product instructions. Structured information means that after analysis, the information can be decomposed into multiple interrelated components, and there is a clear hierarchical structure between the components. Its use and maintenance are managed through a database and have certain operating specifications. For example, in the case where the product instructions are drug instructions, product structured information includes but is not limited to at least one of the following types: generic name of the drug, commonly used name of the drug, molecular formula of the drug, indications of the drug, usage and dosage of the drug, and contraindications of the drug.

[0045] The product relationship network refers to a relationship network constructed based on the association relationship between product structured information in each product manual. The product relationship network consists of nodes and edge relationships between nodes. The nodes in the product relationship network are generated based on the product structured information, and the edge relationships between nodes in the product relationship network are generated based on the association relationship between product structured information.

[0046] In one embodiment, a server for training a model training method (hereinafter referred to as the first server) obtains at least one product manual from a database, and extracts structured information from each product manual to determine at least one type of product structured information included in each product manual. Further, the first server generates nodes in a product relationship network based on the product structured information included in each product manual, and generates edge relationships between nodes in the product relationship network based on the association relationship between the product structured information included in each product manual, thereby constructing a product relationship network based on the edge relationships between nodes.

[0047] S102: Determine any node in the product relationship network as a target node, and determine at least one first-order neighbor node corresponding to the target node as a target neighbor node according to the product relationship network.

[0048] The target node refers to any node in the product relationship network, that is, each node included in the product relationship network can be used as a target node.

[0049] In one embodiment, the first server determines the target node from the product relationship network, and determines at least one first-order neighbor node corresponding to the target node as the target neighbor node based on the edge relationship between the nodes in the product relationship network. Among them, the target neighbor node may include a neighbor node having a direct edge relationship with the target node (directly connected to the target node), and may also include a neighbor node having an indirect edge relationship with the target node (indirectly connected to the target node). It can be understood that the target neighbor node having a direct edge relationship with the target node is a first-order neighbor node of the target node; the target neighbor node having an indirect edge relationship with the target node is an N-order neighbor node of the target node, where N is greater than or equal to two. For example, when N is two, the target neighbor node is a neighbor node having a direct edge relationship with the first-order neighbor node of the target node; when N is three, the target neighbor node is a neighbor node having a direct edge relationship with the second-order neighbor node of the target node, and so on.

[0050] S103 . Update the feature vector corresponding to the target node according to the feature vectors corresponding to each target neighbor node, and generate an updated feature vector corresponding to the target node.

[0051] Among them, the feature vector corresponding to the node is also called embedding, which is a vector representation that can reflect the semantic information of the corresponding node. Since there is a direct or indirect edge relationship between each target neighbor node and the target node, there is a certain association relationship between each target neighbor node and the target node. Therefore, the feature vector corresponding to each target neighbor node can be used to update the feature vector corresponding to the target node to improve the feature information richness of the feature vector corresponding to the target node.

[0052] In one embodiment, the first server uses feature vector generation technology to determine the feature vectors corresponding to each node in the product relationship network, and updates the feature vectors corresponding to the target node according to the mean calculation results of the feature vectors corresponding to each target neighbor node to generate an updated feature vector corresponding to the target node.

[0053] In another embodiment, the first server uses feature vector generation technology to determine the feature vectors corresponding to each node in the product relationship network, and determines important neighbor nodes from each target neighbor node based on the influence weight value of each target neighbor node on the target node, and updates the feature vector corresponding to the target node based on the feature vector corresponding to the important neighbor node to generate an updated feature vector corresponding to the target node.

[0054] In another embodiment, the first server uses feature vector generation technology to determine the feature vectors corresponding to each node in the product relationship network, and updates the feature vectors corresponding to the target node based on the mean calculation results of the feature vectors corresponding to each target neighbor node and the feature vectors corresponding to the important neighbor nodes, thereby generating an updated feature vector corresponding to the target node.

[0055] S104: Use the product structured information corresponding to the target node as a sample query word, and use the updated feature vector as a sample feature vector corresponding to the sample query word, and train the to-be-trained model according to the sample query word and the sample feature vector to generate a feature vector prediction model.

[0056] The feature vector prediction model is used to predict the corresponding target feature vector according to the user's original query word, and the target feature vector is used to query product information from the product manual.

[0057] In one implementation, the first server uses the product structured information corresponding to the target node as a sample query word, uses the updated feature vector as a sample feature vector corresponding to the sample query word, and constructs a training data set according to the sample query word and the sample feature vector.

[0058] During the process of training the model to be trained, the first server uses the sample query words in the training data set as input data for the training of the model to be trained, and uses the sample feature vectors in the training data set as output data for the training of the model to be trained, trains the model to be trained, and generates a feature vector prediction model.

[0059] After the feature vector prediction model is trained, it can be used to query product information. Specifically, the user's original query word is input into the feature vector prediction model, and the feature vector prediction model is used to predict the corresponding target feature vector, and then the product information is queried from the product manual according to the target feature vector. Since the target feature vector contains rich feature information, the accuracy of product information query can be improved.

[0060] The present invention determines at least one type of product structured information included in at least one product description, and constructs a product relationship network according to the product structured information; determines any node from the product relationship network as a target node, and determines at least one first-order neighbor node corresponding to the target node as a target neighbor node according to the product relationship network; updates the feature vector corresponding to the target node according to the feature vectors corresponding to each target neighbor node, and generates an updated feature vector corresponding to the target node; uses the product structured information corresponding to the target node as a sample query word, and uses the updated feature vector as a sample feature vector corresponding to the sample query word, and trains a to-be-trained model according to the sample query word and the sample feature vector, and generates a feature vector prediction model. type; wherein, the feature vector prediction model is used to predict the corresponding target feature vector according to the user's original query term, and the target feature vector is used to query product information from the product manual. Since the updated feature vector corresponding to the target node not only contains its own feature information, but also includes the feature information of its corresponding target neighbor node, the sample feature vector determined according to the updated feature vector contains richer feature information, and further enables the feature vector prediction model trained according to the sample feature vector to predict the target feature vector containing rich feature information according to the original query term. Since the target feature vector has rich feature information, it is ultimately guaranteed that more accurate product information can be queried according to the target feature vector, thereby ensuring user experience.

[0061] Embodiment 2

[0062] Figure 2 This is a flow chart of a model training method provided in the second embodiment of the present invention. This embodiment further expands and optimizes the above embodiment and can be combined with the above optional implementation modes. Figure 2 As shown, the method includes:

[0063] S201. Determine at least one type of product structured information included in at least one drug instruction sheet.

[0064] S202: The generic name of a drug included in any drug instruction sheet is used as a central node, and the product structured information other than the generic name of the drug included in the drug instruction sheet is used as other nodes.

[0065] For example, assuming that the generic name of a drug included in any drug instruction sheet is "A", and the product structured information included in the drug instruction sheet in addition to the generic name of the drug is the common name of the drug "B", the molecular formula of the drug "C", the indication of the drug "D", the usage and dosage of the drug "E" and the contraindications of the drug "F", then the generic name of the drug "A" is taken as the central node, and the common name of the drug "B", the molecular formula of the drug "C", the indication of the drug "D", the usage and dosage of the drug "E" and the contraindications of the drug "F" are taken as other nodes respectively.

[0066] S203, constructing edge relationships between other nodes and the central node, generating a drug relationship sub-network corresponding to the drug instructions, and constructing a drug relationship network according to the drug relationship sub-networks corresponding to each drug instruction as a product relationship network.

[0067] In one embodiment, for the drug instruction sheet, the edge relationship between each other node and the central node is constructed to generate a drug relationship sub-network corresponding to the drug instruction sheet. Further, for other drug instructions other than the drug instruction sheet, the same process is also used to construct the drug relationship sub-networks corresponding to each other drug instruction sheet, and then the drug relationship network is constructed according to the drug relationship sub-networks corresponding to each drug instruction sheet as the product relationship network.

[0068] Continuing with the above example, the construction of the drug relationship subnetwork is explained. For the drug instructions, assume that the other nodes corresponding to the drug common name "B", drug molecular formula "C", drug indication "D", drug usage and dosage "E", and drug contraindications "F" are respectively used as other nodes 1, other nodes 2, other nodes 3, other nodes 4, and other nodes 5, and then construct the edge relationships between other nodes 1, other nodes 2, other nodes 3, other nodes 4, and other nodes 5 and the central node, respectively, to generate the drug relationship subnetwork corresponding to the drug instructions.

[0069] The generic name of a drug included in any drug instruction is taken as the central node, and the product structured information other than the generic name of the drug included in the drug instruction is taken as other nodes; the edge relationship between each other node and the central node is constructed to generate a drug relationship sub-network corresponding to the drug instruction, and a drug relationship network is constructed according to the drug relationship sub-networks corresponding to each drug instruction as a product relationship network, thereby providing a method for constructing a product relationship network when the product instruction is a drug instruction, and due to the uniqueness of the generic name of a drug in a single drug instruction, taking it as the central node to construct the drug relationship sub-network can ensure the accuracy and rationality of the construction of the drug relationship sub-network.

[0070] Optionally, constructing a drug relationship network according to the drug relationship sub-networks corresponding to the drug instructions respectively includes:

[0071] A1. Perform similarity matching on the first-category nodes included in different drug relationship sub-networks.

[0072] Among them, the first type of nodes are the nodes corresponding to the drug indications in each drug relationship sub-network.

[0073] In one embodiment, the first server determines the first type of nodes included in each drug relationship sub-network, and performs similarity matching on the first type of nodes included in different drug relationship sub-networks.

[0074] B1. If the similarity between the first candidate node included in the first drug relationship sub-network and the second candidate node included in the second drug relationship sub-network is greater than a first threshold, an edge relationship between the first candidate node and the second candidate node is constructed to generate a drug relationship network.

[0075] Among them, the first drug relationship subnetwork is any drug relationship subnetwork, the second drug relationship subnetwork is any drug relationship subnetwork except the first drug relationship subnetwork, the first candidate node is any first-class node in the first drug relationship subnetwork, and the second candidate node is any first-class node in the second drug relationship subnetwork.

[0076] In one embodiment, if the first server determines that the first candidate node included in the first drug relationship sub-network and the second candidate node included in the second drug relationship sub-network have a similarity greater than a first threshold, it means that the first candidate node and the second candidate node are the same or highly similar, for example, the first candidate node is "fever" and the second candidate node is also "fever"; for another example, the first candidate node is "fever" and the second candidate node is "fever", etc. At this time, it can be considered that the first candidate node and the second candidate node are the same nodes in terms of semantic information, so an edge relationship between the first candidate node and the second candidate node is constructed to generate a drug relationship network.

[0077] The first-category nodes included in different drug relationship sub-networks are matched for similarity respectively; wherein the first-category nodes are nodes corresponding to drug indications in each drug relationship sub-network; if the similarity between the first candidate node included in the first drug relationship sub-network and the second candidate node included in the second drug relationship sub-network is greater than a first threshold, an edge relationship between the first candidate node and the second candidate node is constructed to generate a drug relationship network, thereby connecting different drug relationship sub-networks using nodes corresponding to the same or similar drug indications, thereby ensuring the integrity and information richness of the drug relationship network.

[0078] Optionally, a drug relationship network is constructed according to the drug relationship sub-networks corresponding to each drug instruction sheet, including:

[0079] A2. Perform similarity matching on the second-category nodes included in different drug relationship sub-networks.

[0080] Among them, the second type of nodes are the nodes corresponding to drug contraindications in each drug relationship sub-network.

[0081] In one embodiment, the first server determines the second type of nodes included in each drug relationship sub-network, and performs similarity matching on the second type of nodes included in different drug relationship sub-networks.

[0082] B2. If the similarity between the third candidate node included in the third drug relationship sub-network and the fourth candidate node included in the fourth drug relationship sub-network is greater than the second threshold, an edge relationship between the third candidate node and the fourth candidate node is constructed to generate a drug relationship network.

[0083] Among them, the third drug relationship subnetwork is any drug relationship subnetwork, the fourth drug relationship subnetwork is any drug relationship subnetwork except the third drug relationship subnetwork, the third candidate node is any second-class node in the third drug relationship subnetwork, and the fourth candidate node is any second-class node in the fourth drug relationship subnetwork.

[0084] In one embodiment, if the first server determines that the third candidate node included in the third drug relationship sub-network and the fourth candidate node included in the fourth drug relationship sub-network have a similarity greater than the second threshold, it means that the third candidate node and the fourth candidate node are the same or highly similar, for example, the third candidate node is "drinking", and the fourth candidate node is also "drinking"; for another example, the third candidate node is "drinking", and the fourth candidate node is "alcoholism", etc. At this time, it can be considered that the third candidate node and the fourth candidate node are the same nodes in terms of semantic information, so the edge relationship between the third candidate node and the fourth candidate node is constructed to generate a drug relationship network.

[0085] The second-category nodes included in different drug relationship sub-networks are matched for similarity respectively; wherein the second-category nodes are nodes corresponding to drug contraindications in each drug relationship sub-network; if the similarity between the third candidate node included in the third drug relationship sub-network and the fourth candidate node included in the fourth drug relationship sub-network is greater than the second threshold, then an edge relationship between the third candidate node and the fourth candidate node is constructed to generate a drug relationship network, thereby connecting different drug relationship sub-networks using the same or similar nodes corresponding to drug contraindications, thereby ensuring the integrity and information richness of the drug relationship network.

[0086] S204: Determine any node in the product relationship network as a target node, and determine at least one first-order neighbor node corresponding to the target node as a target neighbor node according to the product relationship network.

[0087] S205 . Classify each target neighbor node according to the information category of the product structured information corresponding to each target neighbor node to obtain at least one neighbor node category.

[0088] Among them, product structured information of different information categories corresponds to different categories of structured information content. Taking the product instructions as drug instructions as an example, the information categories of the product structured information include but are not limited to the generic name of the drug, the commonly used name of the drug, the molecular formula of the drug, the indications of the drug, the usage and dosage of the drug, and the contraindications of the drug.

[0089] In one implementation, according to the information category of the product structured information corresponding to each target neighbor node, the target neighbor nodes belonging to the same information category are clustered to obtain at least one neighbor node category.

[0090] For example, the information category of the product structured information corresponding to the target neighbor node 1 is category 1, the information category of the product structured information corresponding to the target neighbor node 2 is category 2, the information category of the product structured information corresponding to the target neighbor node 3 is category 1, and the information category of the product structured information corresponding to the target neighbor node 4 is category 2, then the target neighbor node 1 and the target neighbor node 3 are clustered as the same neighbor node category; the target neighbor node 2 and the target neighbor node 4 are clustered as the same neighbor node category.

[0091] S206 , performing mean operation on the feature vectors corresponding to the target neighbor nodes in any neighbor node category to generate a feature mean vector, and determining a category feature vector corresponding to the neighbor node category according to the feature mean vector.

[0092] In one embodiment, the first server uses a gelu function to perform a nonlinear transformation on the feature vectors corresponding to each target neighbor node to strengthen the feature expression of the feature vectors corresponding to each target neighbor node. Further, the first server determines the weighted weight values ​​corresponding to each target neighbor node in each neighbor node category, and performs a weighted mean operation on the feature vectors corresponding to each target neighbor node in each neighbor node category according to the weighted weight values ​​to generate a feature mean vector, and then determines the category feature vector corresponding to each neighbor node category according to the feature mean vector. Among them, the weighted weight values ​​corresponding to each target neighbor node can be iteratively updated according to the continuous training of the model to be trained.

[0093] S207 , performing feature splicing according to the category feature vectors corresponding to the categories of each neighbor node to generate a feature splicing vector, and generating an updated feature vector corresponding to the target node according to the feature splicing vector and the feature vector corresponding to the target node.

[0094] In one embodiment, the first server performs feature splicing on the category feature vectors corresponding to each neighbor node category to generate a feature splicing vector, and performs feature splicing on the feature splicing vector and the feature vector corresponding to the target node to generate an updated feature vector corresponding to the target node.

[0095] By calculating the feature mean vector of the target neighbor nodes in each information category, an updated feature vector corresponding to the target node is generated, thereby capturing the comprehensive feature performance of the target neighbor nodes in each information category and enriching the feature information expression of the updated feature vector corresponding to the target node.

[0096] Optionally, determining a category feature vector corresponding to the category of the neighbor node according to the feature mean vector includes:

[0097] S2061. Target neighbor nodes included in the neighbor node category are taken as auxiliary neighbor nodes, and influence weight values ​​of the auxiliary neighbor nodes on the target node are determined, and important neighbor nodes are determined from the auxiliary neighbor nodes according to the ranking results of the influence weight values.

[0098] The influence weight value reflects the influence of the auxiliary neighbor node on the target node. It can be understood that the greater the influence weight value of any auxiliary neighbor node on the target node, the greater the influence of the auxiliary neighbor node on the target node, or the more important the auxiliary neighbor node is to the target node. The influence weight value can be iteratively updated according to the continuous training of the model to be trained.

[0099] In one embodiment, the first server determines the influence weight value of each auxiliary neighbor node on the target node, and sorts the influence weight values. According to the sorting result of the influence weight values, the auxiliary neighbor node corresponding to the largest influence weight value is regarded as an important neighbor node.

[0100] S2062: Add features according to the feature vectors and feature mean vectors corresponding to the important neighbor nodes to generate a category feature vector corresponding to the category of the neighbor node.

[0101] In one implementation, the first server performs feature addition on a feature vector corresponding to an important neighbor node of the neighbor node category and a feature mean vector corresponding to the neighbor node category to generate a category feature vector corresponding to the neighbor node category.

[0102] By determining the important neighbor nodes with the greatest influence on the target node in each information category, and adding the features according to the feature vectors and feature mean vectors corresponding to the important neighbor nodes, the category feature vector corresponding to the neighbor node category is generated. The category feature vector contains both the comprehensive feature performance of the target neighbor node in each information category and the outstanding feature performance of the important neighbor nodes, further enriching the feature information expression of the updated feature vector corresponding to the target node.

[0103] S208: Use the product structured information corresponding to the target node as a sample query word, and use the updated feature vector as a sample feature vector corresponding to the sample query word, and train the to-be-trained model according to the sample query word and the sample feature vector to generate a feature vector prediction model.

[0104] Embodiment 3

[0105] Figure 3 This is a flowchart of a product information query method provided in the third embodiment of the present invention. This embodiment is applicable to the case where a trained feature vector prediction model is used to query product information. The method can be executed by a product information query device, which can be implemented in the form of hardware and / or software, such as a server. Figure 3 As shown, the method includes:

[0106] S301. Determine an original query word according to a user's query statement, and input the original query word into a feature vector prediction model to predict a target feature vector corresponding to the original query word.

[0107] The feature vector prediction model is trained using the model training method described above.

[0108] In one embodiment, a server for querying product information (hereinafter referred to as the second server) receives a query statement implemented by a user, and performs keyword extraction on the query statement to determine at least one keyword contained in the query statement as an original query word. The second server further inputs the original query word into a deployed feature vector prediction model, uses the feature vector prediction model to perform feature vector prediction on the original query word, and outputs a target feature vector corresponding to the original query word.

[0109] S302: Determine a target standard query word from candidate standard query words according to the target feature vector, and use the target standard query word to query product information from at least one product manual.

[0110] The candidate standard query words are determined according to the product structured information included in the product specification.

[0111] In one embodiment, the second server inputs the target feature vector into the query word prediction model, and the query word prediction model predicts the target standard query word from the candidate standard query words according to the target feature vector. The query word prediction model is trained based on the sample feature vector and the sample query word as training data, that is, the query word prediction model has the ability to predict the corresponding query word according to the feature vector.

[0112] Furthermore, the second server uses an information query algorithm to use the target standard query word as a keyword to perform a product information query in at least one product manual, and obtains product information matching the target standard query word as a recall result and feeds it back to the user.

[0113] On the one hand, since the feature vector prediction model trained by the model training method as described above can predict the target feature vector containing rich feature information, the present invention determines the original query word according to the user's query statement, and inputs the original query word into the feature vector prediction model to predict the target feature vector corresponding to the original query word, determines the target standard query word from the candidate standard query word according to the target feature vector, and uses the target standard query word to query product information from at least one product manual, which can ensure that more comprehensive and accurate target standard query words are determined according to the target feature vector, and further ensure that more comprehensive and accurate product information can be queried according to the target standard query word, thereby ensuring user experience; on the other hand, since the candidate standard query word is set to be determined according to the product structured information included in the product manual, the standardization of the candidate standard query word can be guaranteed, and the accuracy of product information queried according to the target standard query word is further improved.

[0114] Embodiment 4

[0115] Figure 4 This is a schematic diagram of the structure of a model training device provided in Embodiment 4 of the present invention. Figure 4 As shown, the device comprises:

[0116] A product relationship network construction module 41 is used to determine at least one type of product structured information included in at least one product specification, and to construct a product relationship network according to the product structured information;

[0117] A target neighbor node determination module 42 is used to determine any node from the product relationship network as a target node, and determine at least one first-order neighbor node corresponding to the target node as a target neighbor node according to the product relationship network;

[0118] An updated feature vector generating module 43 is used to update the feature vector corresponding to the target node according to the feature vectors corresponding to each of the target neighbor nodes, and generate an updated feature vector corresponding to the target node;

[0119] A model training module 44 is used to use the product structured information corresponding to the target node as a sample query word, and use the updated feature vector as a sample feature vector corresponding to the sample query word, and train a to-be-trained model according to the sample query word and the sample feature vector to generate a feature vector prediction model;

[0120] The feature vector prediction model is used to predict the corresponding target feature vector according to the user's original query word, and the target feature vector is used to query product information from the product manual.

[0121] Optionally, in the case where the product instructions are drug instructions, the product structured information is at least one of a drug generic name, a drug common name, a drug molecular formula, a drug indication, a drug usage and dosage, and a drug contraindication;

[0122] The product relationship network construction module 41 is specifically used for:

[0123] The generic name of the drug included in any of the drug instructions is used as a central node, and the product structured information other than the generic name of the drug included in the drug instructions is used as other nodes;

[0124] Construct edge relationships between each of the other nodes and the central node, generate a drug relationship subnetwork corresponding to the drug instructions, and construct a drug relationship network based on the drug relationship subnetworks corresponding to each of the drug instructions as the product relationship network.

[0125] Optionally, the product relationship network construction module 41 is further configured to:

[0126] Performing similarity matching on the first-type nodes included in different drug relationship sub-networks respectively; wherein the first-type nodes are nodes corresponding to the drug indications in each of the drug relationship sub-networks;

[0127] If the similarity between the first candidate node included in the first drug relationship sub-network and the second candidate node included in the second drug relationship sub-network is greater than a first threshold, constructing an edge relationship between the first candidate node and the second candidate node to generate the drug relationship network;

[0128] Among them, the first drug relationship subnetwork is any of the drug relationship subnetworks, the first drug relationship subnetwork is any of the drug relationship subnetworks except the first drug relationship subnetwork, the first candidate node is any of the first-type nodes in the first drug relationship subnetwork, and the second candidate node is any of the first-type nodes in the second drug relationship subnetwork.

[0129] Optionally, the product relationship network construction module 41 is further configured to:

[0130] Performing similarity matching on the second-type nodes included in different drug relationship sub-networks respectively; wherein the second-type nodes are nodes corresponding to the drug contraindications in each of the drug relationship sub-networks;

[0131] If the similarity between the third candidate node included in the third drug relationship sub-network and the fourth candidate node included in the fourth drug relationship sub-network is greater than the second threshold, constructing an edge relationship between the third candidate node and the fourth candidate node to generate the drug relationship network;

[0132] Among them, the third drug relationship subnetwork is any of the drug relationship subnetworks, the fourth drug relationship subnetwork is any of the drug relationship subnetworks except the third drug relationship subnetwork, the third candidate node is any of the second-type nodes in the third drug relationship subnetwork, and the fourth candidate node is any of the second-type nodes in the fourth drug relationship subnetwork.

[0133] Optionally, the updating feature vector generating module 43 is specifically used for:

[0134] Classify each of the target neighbor nodes according to the information category of the product structured information corresponding to each of the target neighbor nodes to obtain at least one neighbor node category;

[0135] Performing a mean operation on the feature vectors corresponding to the target neighbor nodes in any of the neighbor node categories to generate a feature mean vector, and determining a category feature vector corresponding to the neighbor node category according to the feature mean vector;

[0136] Feature splicing is performed according to the category feature vectors corresponding to each of the neighbor node categories to generate a feature splicing vector, and an updated feature vector corresponding to the target node is generated according to the feature splicing vector and the feature vector corresponding to the target node.

[0137] Optionally, the updating feature vector generating module 43 is further configured to:

[0138] The target neighbor node included in the neighbor node category is used as an auxiliary neighbor node, and an influence weight value of each of the auxiliary neighbor nodes on the target node is determined, and an important neighbor node is determined from each of the auxiliary neighbor nodes according to a ranking result of the influence weight values;

[0139] Features are added according to the feature vector corresponding to the important neighbor node and the feature mean vector to generate a category feature vector corresponding to the category of the neighbor node.

[0140] The model training device provided in the embodiment of the present invention can execute the model training method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0141] Embodiment 5

[0142] Figure 5This is a structural diagram of a product information query device provided in Embodiment 5 of the present invention.

[0143] like Figure 5 As shown, the device comprises:

[0144] The feature vector prediction module 51 is used to determine the original query word according to the user's query sentence, and input the original query word into the feature vector prediction model to predict the target feature vector corresponding to the original query word; wherein the feature vector prediction model is trained by the model training method as described above;

[0145] The product information query module 52 is used to determine a target standard query word from candidate standard query words according to the target feature vector, and use the target standard query word to query product information from at least one product specification; wherein the candidate standard query word is determined according to the product structured information included in the product specification.

[0146] The product information query device provided in the embodiment of the present invention can execute the product information query method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0147] Embodiment 6

[0148] Figure 6 A schematic diagram of the structure of an electronic device 60 that can be used to implement an embodiment of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0149] like Figure 6As shown, the electronic device 60 includes at least one processor 61, and a memory connected to the at least one processor 61, such as a read-only memory (ROM) 62, a random access memory (RAM) 63, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 61 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 62 or the computer program loaded from the storage unit 68 to the random access memory (RAM) 63. In the RAM 63, various programs and data required for the operation of the electronic device 60 can also be stored. The processor 61, the ROM 62, and the RAM 63 are connected to each other via a bus 64. An input / output (I / O) interface 65 is also connected to the bus 64.

[0150] A number of components in the electronic device 60 are connected to the I / O interface 65, including: an input unit 66, such as a keyboard, a mouse, etc.; an output unit 67, such as various types of displays, speakers, etc.; a storage unit 68, such as a disk, an optical disk, etc.; and a communication unit 69, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 69 allows the electronic device 60 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0151] The processor 61 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 61 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The processor 61 executes the various methods and processes described above, such as a model training method and / or a product information query method.

[0152] In some embodiments, the model training method and / or the product information query method may be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 68. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 60 via the ROM 62 and / or the communication unit 69. When the computer program is loaded into the RAM 63 and executed by the processor 61, one or more steps of the model training method and / or the product information query method described above may be performed. Alternatively, in other embodiments, the processor 61 may be configured to execute the model training method and / or the product information query method in any other appropriate manner (e.g., by means of firmware).

[0153] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0154] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0155] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0156] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0157] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0158] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.

[0159] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0160] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A model training method, characterized in that: include: Determining at least one type of product structured information included in at least one product specification, and constructing a product relationship network based on the product structured information; Determine any node from the product relationship network as a target node, and determine at least one first-order neighbor node corresponding to the target node as a target neighbor node according to the product relationship network; According to the feature vectors corresponding to each of the target neighbor nodes, the feature vector corresponding to the target node is updated to generate an updated feature vector corresponding to the target node; Taking the product structured information corresponding to the target node as a sample query word, and taking the updated feature vector as a sample feature vector corresponding to the sample query word, and training a to-be-trained model according to the sample query word and the sample feature vector to generate a feature vector prediction model; The feature vector prediction model is used to predict the corresponding target feature vector according to the user's original query word, and the target feature vector is used to query product information from the product manual.

2. The method according to claim 1, characterized in that In the case where the product instructions are drug instructions, the product structured information is at least one of the generic name of the drug, the commonly used name of the drug, the molecular formula of the drug, the indications of the drug, the usage and dosage of the drug, and the contraindications of the drug; The step of constructing a product relationship network according to the product structured information includes: The generic name of the drug included in any of the drug instructions is used as a central node, and the product structured information other than the generic name of the drug included in the drug instructions is used as other nodes; Construct edge relationships between each of the other nodes and the central node, generate a drug relationship subnetwork corresponding to the drug instructions, and construct a drug relationship network based on the drug relationship subnetworks corresponding to each of the drug instructions as the product relationship network.

3. The method according to claim 2, characterized in that The step of constructing a drug relationship network according to the drug relationship sub-networks corresponding to the drug instructions includes: Performing similarity matching on the first-type nodes included in different drug relationship sub-networks respectively; wherein the first-type nodes are nodes corresponding to the drug indications in each of the drug relationship sub-networks; If the similarity between the first candidate node included in the first drug relationship sub-network and the second candidate node included in the second drug relationship sub-network is greater than a first threshold, constructing an edge relationship between the first candidate node and the second candidate node to generate the drug relationship network; Among them, the first drug relationship subnetwork is any of the drug relationship subnetworks, the first drug relationship subnetwork is any of the drug relationship subnetworks except the first drug relationship subnetwork, the first candidate node is any of the first-type nodes in the first drug relationship subnetwork, and the second candidate node is any of the first-type nodes in the second drug relationship subnetwork.

4. The method according to claim 2, characterized in that: The step of constructing a drug relationship network according to the drug relationship sub-networks corresponding to the drug instructions includes: Performing similarity matching on the second-type nodes included in different drug relationship sub-networks respectively; wherein the second-type nodes are nodes corresponding to the drug contraindications in each of the drug relationship sub-networks; If the similarity between the third candidate node included in the third drug relationship sub-network and the fourth candidate node included in the fourth drug relationship sub-network is greater than the second threshold, constructing an edge relationship between the third candidate node and the fourth candidate node to generate the drug relationship network; Among them, the third drug relationship subnetwork is any of the drug relationship subnetworks, the fourth drug relationship subnetwork is any of the drug relationship subnetworks except the third drug relationship subnetwork, the third candidate node is any of the second-type nodes in the third drug relationship subnetwork, and the fourth candidate node is any of the second-type nodes in the fourth drug relationship subnetwork.

5. The method according to claim 1, characterized in that The updating of the feature vector corresponding to the target node according to the feature vectors corresponding to each of the target neighbor nodes to generate an updated feature vector corresponding to the target node includes: Classify each of the target neighbor nodes according to the information category of the product structured information corresponding to each of the target neighbor nodes to obtain at least one neighbor node category; Performing a mean operation on the feature vectors corresponding to the target neighbor nodes in any of the neighbor node categories to generate a feature mean vector, and determining a category feature vector corresponding to the neighbor node category according to the feature mean vector; Feature splicing is performed according to the category feature vectors corresponding to each of the neighbor node categories to generate a feature splicing vector, and an updated feature vector corresponding to the target node is generated according to the feature splicing vector and the feature vector corresponding to the target node.

6. The method according to claim 5, characterized in that The step of determining the category feature vector corresponding to the category of the neighbor node according to the feature mean vector includes: The target neighbor node included in the neighbor node category is used as an auxiliary neighbor node, and an influence weight value of each of the auxiliary neighbor nodes on the target node is determined, and an important neighbor node is determined from each of the auxiliary neighbor nodes according to a ranking result of the influence weight values; Features are added according to the feature vector corresponding to the important neighbor node and the feature mean vector to generate a category feature vector corresponding to the category of the neighbor node.

7. A product information query method, characterized in that: include: Determine an original query word according to the user's query statement, and input the original query word into a feature vector prediction model to predict a target feature vector corresponding to the original query word; wherein the feature vector prediction model is trained using the model training method according to any one of claims 1 to 6; A target standard query word is determined from candidate standard query words according to the target feature vector, and the target standard query word is used to query product information from at least one product specification; wherein the candidate standard query word is determined according to product structured information included in the product specification.

8. A model training device, characterized in that: include: A product relationship network construction module, used to determine at least one type of product structured information included in at least one product specification, and to construct a product relationship network according to the product structured information; A target neighbor node determination module, used to determine any node from the product relationship network as a target node, and determine at least one first-order neighbor node corresponding to the target node as a target neighbor node according to the product relationship network; An updated feature vector generation module is used to update the feature vector corresponding to the target node according to the feature vectors corresponding to each of the target neighbor nodes, and generate an updated feature vector corresponding to the target node; A model training module, used to use the product structured information corresponding to the target node as a sample query word, and the updated feature vector as a sample feature vector corresponding to the sample query word, and train a to-be-trained model according to the sample query word and the sample feature vector to generate a feature vector prediction model; The feature vector prediction model is used to predict the corresponding target feature vector according to the user's original query word, and the target feature vector is used to query product information from the product manual.

9. A product information query device, characterized in that: include: A feature vector prediction module, used to determine an original query word according to a user's query statement, and input the original query word into a feature vector prediction model to predict a target feature vector corresponding to the original query word; wherein the feature vector prediction model is trained using the model training method according to any one of claims 1 to 6; A product information query module is used to determine a target standard query word from candidate standard query words according to the target feature vector, and use the target standard query word to query product information from at least one product specification; wherein the candidate standard query word is determined according to the product structured information included in the product specification.

10. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the model training method described in any one of claims 1 to 6, and / or implement the product information query method described in claim 7.

11. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to execute the model training method described in any one of claims 1 to 6, and / or to implement the product information query method described in claim 7.

12. A computer program product, comprising a computer program, which, when executed by a processor, implements the model training method according to any one of claims 1 to 6, and / or implements the product information query method according to claim 7.

Images