WO2022095893A1 - 生成推荐信息的方法、装置 - Google Patents

生成推荐信息的方法、装置 Download PDF

Info

Publication number
WO2022095893A1
WO2022095893A1 PCT/CN2021/128399 CN2021128399W WO2022095893A1 WO 2022095893 A1 WO2022095893 A1 WO 2022095893A1 CN 2021128399 W CN2021128399 W CN 2021128399W WO 2022095893 A1 WO2022095893 A1 WO 2022095893A1
Authority
WO
WIPO (PCT)
Prior art keywords
information
feature
recommendation
historical
user
Prior art date
Application number
PCT/CN2021/128399
Other languages
English (en)
French (fr)
Inventor
刘宗节
王永鹏
Original Assignee
北京京东拓先科技有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 北京京东拓先科技有限公司 filed Critical 北京京东拓先科技有限公司
Publication of WO2022095893A1 publication Critical patent/WO2022095893A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9535—Search customisation based on user profiles and personalisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/33—Querying
    • G06F16/332—Query formulation
    • G06F16/3329—Natural language query formulation or dialogue systems
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/33—Querying
    • G06F16/3331—Query processing
    • G06F16/334—Query execution
    • G06F16/3344—Query execution using natural language analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/70—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for mining of medical data, e.g. analysing previous cases of other patients

Definitions

  • the present application relates to the field of artificial intelligence, in particular to the field of natural language processing technology and the field of big data technology, and in particular to a method, apparatus, electronic device, and computer-readable storage medium for generating recommendation information.
  • the triage system When a patient conducts a consultation through an Internet hospital, the triage system will simply interact with the patient to obtain the patient's case information in advance, and then the triage system will perform natural language processing based on the patient's case information and conduct intelligent sub-departments to replace the traditional The rule matching has brought many benefits.
  • a language recognition neural network is usually used to identify the current description of the user, and according to the obtained identification result, the department label is matched to complete the department assignment and recommendation.
  • the present application provides a method, apparatus, electronic device and storage medium for generating recommendation information.
  • an embodiment of the present application provides a method for generating recommendation information, including: in response to determining that the user is not accessing the user for the first time, acquiring current description information input by the user and historical description information of the user; The information is input into the historical feature matching model for processing to obtain historical feature information; the current feature information corresponding to the current description information is queried; the historical feature information and the current feature information are spliced to generate recommended feature information; the predetermined recommendation feature is used to identify the neural The network identifies the recommendation feature information, obtains first recommendation information, and sends the first recommendation information to the user.
  • the method further includes: in response to the user being a first-time visiting user, obtaining the current description information input by the user; querying the current feature information corresponding to the current description information; identifying the current feature information by using the recommendation feature recognition neural network , obtain the second recommendation information, and send the second recommendation information to the user.
  • inputting the historical description information into the historical feature matching model for processing, and obtaining the historical feature information includes: acquiring the user description information and the recommendation result description information in the historical description information of the user; classifying the user description information After normalization, it is input to the first deep learning neural network layer constituting the historical feature matching model for processing to generate first feature information; the recommendation result description information is input into the multi-layer bidirectional transformer coding constituting the historical feature matching model layer to generate second feature information; input the first feature information and the second feature information into the second deep learning network layer that constitutes the historical feature matching model for splicing to obtain the historical feature information.
  • querying the current feature information corresponding to the current description information includes: inputting the current description information to a multi-layer bidirectional transformer encoder layer for outputting the current feature information, and querying the current feature information corresponding to the current description information.
  • obtaining historical feature information and current feature information, and splicing to generate recommended feature information includes: respectively obtaining a historical feature vector and a current feature vector corresponding to the historical feature information and the current feature information;
  • the current feature vector is input to the splicing deep learning neural network layer for splicing, and the recommended feature vector is generated by splicing as the recommended feature information; wherein, the number of dimensions of the recommended feature vector is based on the addition of the number of dimensions of the historical feature vector and the number of dimensions of the current feature vector and get.
  • using a predetermined recommendation feature recognition neural network to identify recommendation feature information, obtaining first recommendation information, and sending the first recommendation information to the user includes: using a predetermined recommendation feature recognition neural network to identify the recommendation feature information, and iteratively generate multiple recommendation information; wherein, the loss function of the recommendation feature recognition neural network is a cross entropy function; in response to determining that the loss function converges, the iterative identification is stopped, and the finally obtained recommendation information is determined as the first recommendation information, and send the first recommendation information to the user.
  • an embodiment of the present application provides an apparatus for generating recommendation information, comprising: a description information obtaining unit, configured to obtain the current description information input by the user and the user's current description information in response to determining that the user is not accessing the user for the first time.
  • historical description information e.g., historical description information
  • a historical feature matching unit e.g., historical feature matching unit
  • a current feature query unit e.g., current feature query unit
  • the recommendation feature generating unit is configured to splicing the historical feature information and the current feature information to generate recommendation feature information
  • the recommendation information sending unit is configured to use a predetermined recommendation feature recognition neural network to identify the recommendation feature information, and obtain the first recommendation feature information. recommending information, and sending the first recommending information to the user.
  • the description information obtaining unit is further configured to, in response to the user being a first-time visiting user, obtaining the current description information input by the user includes; the recommendation feature generation unit is further configured to query the current description information corresponding current feature information; the recommendation information sending unit is further configured to use the recommendation feature recognition neural network to identify the current feature information, obtain second recommendation information, and send the second recommendation information to the user.
  • the historical feature matching unit includes: a historical information acquisition subunit, configured to acquire user description information and recommendation result description information in the user's historical description information; a first feature information generation subunit, configured After the user description information is normalized, it is input to the first deep learning neural network layer constituting the historical feature matching model for processing to generate first feature information; the second feature information generating subunit is configured to The recommendation result description information is input into the multi-layer bidirectional transformer encoder layer constituting the historical feature matching model, and the second feature information is generated; the historical feature information generating subunit is configured to the first feature information and the second feature information. Input to the second deep learning network layer constituting the historical feature matching model for splicing to obtain the historical feature information.
  • the current feature query unit is further configured to: input the current description information to the multi-layer bidirectional transformer encoder layer for outputting the current feature information, and query the current feature information corresponding to the current description information.
  • the recommendation feature generation unit includes: a feature vector generation subunit, configured to obtain historical feature vectors and current feature vectors corresponding to the historical feature information and the current feature information, respectively; a recommendation vector splicing subunit, configured by be configured to input the historical feature vector and the current feature vector to the splicing deep learning neural network layer for splicing, and splicing to generate a recommended feature vector as the recommended feature information; wherein, the number of dimensions of the recommended feature vector is based on the number of dimensions of the historical feature vector It is obtained by summing with the number of dimensions of the current feature vector.
  • the recommendation information sending unit includes: a recommendation information generating subunit, configured to identify the recommendation feature information by using a predetermined recommendation feature recognition neural network, and iteratively generate a plurality of recommendation information; wherein the recommendation feature identification
  • the loss function of the neural network is a cross-entropy function
  • the recommendation information determination subunit is configured to stop the iterative identification in response to determining that the loss function converges, and determine the finally obtained recommendation information as the first recommendation information
  • the recommendation information sending subunit a unit configured to send the first recommendation information to the user.
  • an embodiment of the present application provides an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a memory that can be executed by the at least one processor Instructions, which are executed by the at least one processor, so that the at least one processor can execute the method for generating recommendation information described in any implementation manner.
  • embodiments of the present application provide a non-transitory computer-readable storage medium storing computer instructions, including: the computer instructions are used to cause the computer to execute the method for generating recommendation information described in any implementation manner.
  • the present application obtains the current description information input by the user and the historical description information of the user, inputs the historical description information into the historical feature matching model for processing, obtains historical feature information, and queries the The current feature information corresponding to the current description information, the historical feature information and the current feature information are spliced together to generate recommended feature information, and then a predetermined recommendation feature recognition neural network is used to identify the recommended feature information, obtain the first recommendation information, and send the recommended feature information.
  • the first recommendation information is given to the user, and the recommendation information is determined in combination with the user's historical description information, so as to provide the user with the recommendation information more accurately, and improve the quality of the recommendation information.
  • FIG. 1 is an exemplary system architecture to which embodiments of the present application may be applied;
  • FIG. 2 is a flowchart of an embodiment of a method for generating recommendation information according to the present application
  • FIG. 3 is a flow chart of an implementation of obtaining historical feature information in the method for generating recommendation information according to the present application
  • FIG. 4 is a schematic structural diagram of an embodiment of an apparatus for generating recommendation information according to the present application.
  • FIG. 5 is a block diagram of an electronic device suitable for implementing the method for generating recommendation information according to an embodiment of the present application.
  • FIG. 1 illustrates an exemplary system architecture 100 to which embodiments of the method, apparatus, electronic device, and computer-readable storage medium for generating recommendation information of the present application may be applied.
  • the system architecture 100 may include terminal devices 101 , 102 , and 103 , a network 104 and a server 105 .
  • the network 104 is a medium used to provide a communication link between the terminal devices 101 , 102 , 103 and the server 105 .
  • the network 104 may include various connection types, such as wired, wireless communication links, or fiber optic cables, among others.
  • the user can use the terminal devices 101, 102, 103 to interact with the server 105 through the network 104, so as to achieve the purpose of sending the user's input information and the like.
  • the terminal devices 101 , 102 , and 103 may be installed with applications related to receiving scheduling instructions and recommendation information, such as service reservation applications, scene query applications, and triage guidance applications.
  • the terminal devices 101, 102, and 103 may be hardware or software. In the case of hardware, it can be various electronic devices with display screens, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, and the like.
  • the terminal devices 101, 102, and 103 are software, they can be installed in the electronic devices listed above. It can be implemented as a plurality of software or software modules (such as sending user input information, etc.), or can be implemented as a single software or software module. There is no specific limitation here.
  • the server 105 may be a server that provides various services, for example, a server that provides push information for the terminal devices 101 , 102 , and 103 .
  • a server that provides push information for the terminal devices 101 , 102 , and 103 .
  • obtain the current description information input by the user and the historical description information of the user input the historical description information into the historical feature matching model for processing, obtain historical feature information, and query the current description
  • the current feature information corresponding to the information, splicing the historical feature information and the current feature information to generate recommended feature information; using a predetermined recommendation feature recognition neural network to identify the recommended feature information, obtain the first recommendation information and send the first recommendation information to the user.
  • the method for generating recommendation information provided by the embodiments of the present application is generally performed by the server 105 , and accordingly, the device for generating recommendation information is generally set in the server 105 .
  • the server may be hardware or software.
  • the server can be implemented as a distributed server cluster composed of multiple servers, or can be implemented as a single server.
  • the server is software, it may be implemented as multiple software or software modules for providing distributed services, or may be implemented as a single software or software module. There is no specific limitation here.
  • the method for generating recommendation information may also be executed by the terminal devices 101 , 102 , and 103 , and correspondingly, the apparatus for generating recommendation information may also be provided in the terminal devices 101 , 102 , and 103 .
  • the example system architecture 100 may also not include the server 105 and the network 104 .
  • terminal devices, networks and servers in FIG. 1 are merely illustrative. There can be any number of terminal devices, networks and servers according to implementation needs.
  • the method for generating recommendation information includes the following steps:
  • Step 201 in response to determining that the user is not accessing the user for the first time, obtain current description information input by the user and historical description information of the user.
  • the execution body of the method for generating recommendation information (for example, the server 105 shown in FIG. 1 ) will first determine whether the user who receives the push information is a user who accesses the push information for the first time. The judgment can be made by means of the user's historical access records, or the user can be verified by means of human-computer interaction, so as to assist the judgment of the above-mentioned execution subject.
  • the current description information input by the user is obtained.
  • the current description information refers to the description information provided by the user this time in order to obtain the corresponding matching recommendation information.
  • the current description information It is input by the user through the human-computer interaction device (for example, the terminal devices 101, 102, 103 shown in Users can complete the acquisition of recommended information through the human-computer interaction equipment they use.
  • the current description information can be the disease information, symptom information, etc. that the user wants to diagnose this time.
  • the recommendation information is used as the food recommendation information
  • the current description information can be The description information may be information such as the taste and taboo information of the dish that the user wishes to taste.
  • the user's historical description information can be obtained simultaneously or sequentially.
  • the historical description information can be used to assign results, symptoms, case information, and diagnosis results to the user's historical department.
  • the execution body collects and stores the user's information for subsequent use.
  • the same historical description information can also be input by the user or saved locally in the human-computer interaction device used by the user.
  • the human-computer interaction device acquires the corresponding historical description information, which is not limited in this application.
  • Step 202 Input the historical description information into the historical feature matching model for processing to obtain historical feature information.
  • the historical description information is input into the historical feature matching model for processing.
  • the historical feature matching model is usually a model that collects and stores the history of a large number of different users. After inputting all or part of the user's historical description information, the historical feature matching model can obtain more comprehensive historical description information of the user, and perform corresponding connection and compression to generate as much as possible containing the user's historical information. Historical feature information, so that the historical feature information can be subsequently used to represent the user's historical description information and historical recommendation information, and to complete the user's behavior analysis based on historical data.
  • the user's historical description information and historical recommendation information generation results are stored in the above-mentioned execution body, and the above-mentioned execution body can correspondingly generate a historical database to record these contents, so as to facilitate the generation of This historical feature matches the model.
  • Step 203 query the current feature information corresponding to the current description information.
  • the corresponding feature extraction can be performed on the current description information by means of feature extraction, feature recognition neural network, etc., so as to obtain the information representing the behavior that the user expects to be recommended this time.
  • Current feature information taking the application scenarios of department classification and consultation as an example, the current description information of the user can be the specific symptom information that the user wishes to diagnose and solve this time, the information of the department the user subjectively expects to go to, and the user's special request information for the selection of the department, etc. information.
  • Step 204 splicing the historical feature information and the current feature information to generate recommended feature information.
  • a suitable splicing form can be selected in combination with the representation form of the feature information to splicing the historical feature information and the current feature information, such as historical feature information.
  • the quantity sets can be combined to obtain the recommended feature information set, or when both the historical feature information and the current feature information are in the form of vectors, the historical feature information can be based on the historical feature information.
  • overlay Splicing for example, the number of dimensions of the historical feature information is A, and the number of dimensions of the current feature information is also A, then the historical feature information and the content of the corresponding dimension of the current feature information are superimposed
  • different constraint rules can also be pre-determined to adjust the form of the obtained recommendation information, for example, when the historical feature information and the current feature information are both expressed in vector form.
  • the recommended information can be pre-determined in the form of a feature vector with a preset number of dimensions.
  • Step 205 using a predetermined recommendation feature recognition neural network to identify the recommendation feature information, obtain first recommendation information, and send the first recommendation information to the user.
  • the predetermined recommendation feature recognition neural network is a recommendation feature neural network determined in advance according to the form of the obtained recommendation feature information, and the recommendation feature information in the historical data may be used as input in advance to obtain the first recommendation feature.
  • the information is used as the output to train the recommendation feature neural network, so that after the recommendation feature information obtained in the above step 204 is obtained, the recommendation feature recognition neural network can be used to identify the recommendation feature information to obtain the final recommendation information.
  • the current description information input by the user and the historical description information of the user are obtained, and the historical description information is input into the historical feature matching model.
  • obtain the first recommendation information send the first recommendation information to the user, determine the recommendation information in combination with the user's historical description information, provide the user with the recommendation information more accurately, and improve the quality of the recommendation information.
  • the method further includes: in response to the user being a first-time visiting user, obtaining the current description information input by the user; querying the current feature information corresponding to the current description information; using the recommendation feature to identify neural
  • the network identifies the current feature information, obtains second recommendation information, and sends the second recommendation information to the user.
  • the current description input by the user is directly obtained, and the method in the above step 203 is also used to obtain the user.
  • the corresponding current feature information because the recommended feature information is spliced from the current feature information and the historical feature information, if there is no historical feature information, it is equivalent to the content of the current feature information and the content of the recommended feature information are the same.
  • the recommendation feature recognition neural network in the above step 205 is also used to identify the recommendation feature to obtain the second recommendation information, and send the second recommendation information to the user to complete the information push work, so as to ensure that when there is no historical description information , that is, the recommendation work when the user visits the user for the first time, avoiding the problem that the recommendation information cannot be generated due to the lack of historical data.
  • a process 300 of processing historical description information by using a historical feature matching model to obtain historical feature information is shown, which specifically includes:
  • Step 301 Obtain user description information and recommendation result description information in the user's historical description information.
  • the user's historical description information usually has a large amount of information content. Therefore, in order to facilitate identification, these information can be divided into two categories, namely user description information and recommendation result description information.
  • user description information is the user's gender, age, days since the last visit, the department of the last visit, the department of the doctor who consulted the last time, the evaluation of the doctor who consulted the last time, whether to purchase medicines or not.
  • the recommended result description information can be the user's latest case description information, symptom information, diagnosis information to issue detailed information of medicines, etc.
  • Step 302 the user description information is normalized and then input to the first deep learning neural network layer constituting the historical feature matching model for processing to generate first feature information.
  • the user description information obtained in the above step 301 is processed by methods such as maximum and minimum normalization and semantic normalization. Normalized processing to obtain a standard processing result that can be identified and extracted by the first deep learning neural network, deep learning neural network (Deep Neural Networks, DNN for short), and then input the result into the first part of the historical feature matching model.
  • a deep learning neural network a deep learning neural network (Deep Neural Networks, DNN for short) can realize the identification and extraction of the processing result, so as to obtain the first feature information used to represent the user description information.
  • the normalized processing result obtained has a total of 15 dimensions
  • the input DNN layer which contains two hidden layers: the first A hidden layer has 60 neurons, and this layer uses a linear rectification activation function (Rectified Linear Unit, referred to as Relu); the second hidden layer has 30 neurons, the activation function is Relu; the number of dimensions of the output layer is 5), the output The layer outputs the first feature information with 5 dimensions.
  • Step 303 Input the recommendation result description information into the multi-layer bidirectional transformer encoder layer constituting the historical feature matching model to generate second feature information.
  • the recommendation result description information is input into the multi-layer Bidirectional Encoder Representations from Transformers constituting the historical feature matching model. , referred to as Bert) for processing to obtain the second feature information used to represent the description information of the recommendation result.
  • the user's latest case description, symptoms and drug details are spliced together as the input of the Bert layer, in order to learn the user's last case characteristics. According to the actual situation analysis, the user's condition has a certain correlation in the short term.
  • the final output dimension of the Bert layer is 10 second feature information.
  • Step 304 Input the first feature information and the second feature information into a second deep learning network layer constituting the historical feature matching model for splicing to obtain the historical feature information.
  • the first feature information and the second feature information obtained in the above steps 302 and 303 are input into the second deep learning network layer for splicing, so as to obtain the final historical feature information.
  • the related content of the first feature information representing the user description information and the second feature information used to represent the recommendation result description information.
  • the outputs of the above-mentioned DNN and Bert models are spliced, that is, the first feature information and the second feature information are spliced, on the one hand, joint training and learning are performed, and the combined features of the two models are abstractly extracted; Some features are compressed, and historical feature information with a dimension of 5 is output to reduce the feature dimension and improve the running performance.
  • This part of the DNN model has three hidden layers: the first hidden layer has 200 neurons, which uses the Relu activation function; the second hidden layer has 100 neurons, which also uses the Relu activation function; the third hidden layer has 60 neurons The neuron also uses the Relu activation function; the output dimension is 5 historical feature information as the user's historical feature information.
  • the historical feature information of the user's historical description information is determined based on the user description information and the recommendation result information through the historical feature matching model, so that the historical feature information can be used to reflect the user's situation more comprehensively.
  • the user's historical data is used to determine the recommendation information, so as to solve the problem in the prior art that high-quality recommendation information cannot be generated due to too little input information of the user, and the content can be accurately recommended for the user.
  • querying the current feature information corresponding to the current description information includes: inputting the current description information into a multi-layer bidirectional transformer encoder layer for outputting the current feature information, and querying the current description information Corresponding current feature information.
  • the recommendation result description information is input into another Bert used for outputting the current feature information, and processing is performed to obtain the current feature information corresponding to the current description information, so as to It realizes that the current description information is processed into the current feature information in a form similar to the historical feature information, which facilitates the splicing of the historical feature information and the current feature information to generate the recommended feature information.
  • acquiring the historical feature information and the current feature information, and splicing to generate the recommended feature information includes: respectively acquiring a historical feature vector and a current feature vector corresponding to the historical feature information and the current feature information ; Input the historical feature vector and the current feature vector to the splicing deep learning neural network layer for splicing, and the splicing generates a recommended feature vector as the recommended feature information; wherein, the number of dimensions of the recommended feature vector is based on the number of dimensions of the historical feature vector and The sum of the number of dimensions of the current feature vector is obtained.
  • the vector is used as the representation form of the historical feature information and the current feature information, and then the historical feature vector and the current feature vector corresponding to the historical feature information and the current feature information are obtained respectively, and then the historical feature vector and the current feature vector.
  • Directly splicing that is, the number of dimensions of the recommended feature vector obtained as the recommended feature information is the sum of the number of dimensions of the historical feature vector and the number of dimensions of the current feature vector.
  • the recommended feature vectors of all historical feature vectors and current feature vector information balance the relationship between the generation efficiency and quality of recommended feature information.
  • the DNN layer is used to splicing the obtained first feature information represented by a 5-dimensional vector and the second feature information represented by a 10-dimensional vector to obtain feature information represented by a 15-dimensional vector.
  • using a predetermined recommendation feature recognition neural network to identify the recommendation feature information, obtaining first recommendation information, and sending the first recommendation information to the user includes: using a predetermined recommendation feature The feature recognition neural network recognizes the recommendation feature information, and iteratively generates a plurality of recommendation information; wherein, the loss function of the recommendation feature recognition neural network is a cross entropy function; in response to determining that the loss function converges, the iterative recognition is stopped, and the final obtained The recommendation information is determined to be the first recommendation information, and the first recommendation information is sent to the user.
  • the recommendation feature identification network After obtaining the recommended feature information, iteratively recognizes the recommended feature by using a predetermined recommended feature recognition network using the cross-entropy function as the loss function in an iterative manner, and stops the iteration in response to determining that the loss function has converged. Identify, send the finally obtained recommendation information as the first recommendation information to the user, and improve the quality of the first recommendation information output by the recommendation feature identification network in an iterative identification manner.
  • this application also provides a specific implementation scheme in combination with a specific application scenario.
  • the current description information "continuous diarrhea from last night to this morning" entered by the user u1 who is not accessing for the first time through the human-computer interaction device used by himself is sent to the executor of the method for generating the recommendation information (recommendation for short). executive body).
  • the recommending subject determines that the user u1 is a non-first-time visitor, and obtains the current description information "continuous diarrhea from last night to this morning" sent by the user u1, and the historical description information obtained from the local area is "gender male, who has visited the internal medicine department, and recently came here.
  • the symptom description information was continuous diarrhea, stomach pains, the diagnosis was superficial gastritis and gastric medicine was prescribed.”
  • the user description information in the historical description information "gender male, has been to the internal medicine department” and the recommended result description information "the most recent symptom description information is continuous diarrhea, stomach pains, the diagnosis result is superficial gastritis and the stomach is prescribed.
  • “Medicine” is input to the first deep learning neural network layer constituting the historical feature matching model and the multi-layer bidirectional transformer encoder layer constituting the historical feature matching model to obtain first feature information and second feature information.
  • the first feature information and the second feature information are input into the second deep learning network layer constituting the historical feature matching model for splicing, and a historical feature vector with a dimension number of 5 is obtained as the historical feature information.
  • the historical feature information and the current feature information are spliced to generate recommended feature information represented by a recommendation feature vector with a dimension of 10, and a predetermined recommendation feature recognition neural network is used to identify the recommended feature information to obtain the first recommendation information. , and send the first recommendation information to the user.
  • the current description information input by the user and the historical description information of the user are input into the historical feature matching model for processing, and the result is obtained Historical feature information, query the current feature information corresponding to the current description information, splicing the historical feature information and the current feature information to generate recommended feature information, and then use a predetermined recommendation feature recognition neural network to identify the recommended feature information, and obtain the first feature information.
  • recommendation information and send the first recommendation information to the user, determine the recommendation information in combination with the user's historical description information, provide the user with the recommendation information more accurately, and improve the quality of the recommendation information.
  • the apparatus 400 for generating recommendation information in this embodiment may include: a description information obtaining unit 401, configured to obtain the current description information input by the user and the history of the user in response to determining that the user is not accessing the user for the first time description information; the historical feature matching unit 402 is configured to input the historical description information into the historical feature matching model for processing to obtain historical feature information; the current feature query unit 403 is configured to query the current feature information corresponding to the current description information The recommended feature generation unit 404 is configured to splicing the historical feature information and the current feature information to generate the recommended feature information; the recommended information sending unit 405 is configured to identify the recommended feature information using a predetermined recommended feature recognition neural network, Obtain the first recommendation information and send the first recommendation information to the user.
  • a description information obtaining unit 401 configured to obtain the current description information input by the user and the history of the user in response to determining that the user is not accessing the user for the first time description information
  • the historical feature matching unit 402 is configured to input the historical description information into the historical feature
  • the above-mentioned apparatus for generating recommendation information further includes: the description information obtaining unit 401 is further configured to, in response to the user being a first-time visiting user, obtain the current description information input by the user Including; the recommendation feature generating unit 404 is further configured to query the current feature information corresponding to the current description information; the recommendation information sending unit 405 is further configured to use the recommendation feature recognition neural network to identify the current feature information, and obtain the second recommending information, and sending the second recommending information to the user.
  • the historical feature matching unit includes: a historical information acquisition subunit configured to acquire user description information and recommendation result description information in the user's historical description information; first feature information The generating subunit is configured to normalize the user description information, and then input it to the first deep learning neural network layer constituting the historical feature matching model for processing to generate the first feature information; the second feature information generates a subunit. a unit configured to input the recommendation result description information into a multi-layer bidirectional transformer encoder layer constituting the historical feature matching model, to generate second feature information; a historical feature information generating subunit, configured to generate the first feature The information and the second feature information are input to the second deep learning network layer that constitutes the historical feature matching model for splicing to obtain the historical feature information.
  • the push information sending unit is further configured to: use a probability graph model to sort the medical state entities, and select a preset number of the medical state entities according to the sorting result to generate a push information set ; Send the push collection to the user.
  • the current feature query unit 403 is further configured to: input the current description information to the multi-layer bidirectional transformer encoder layer for outputting the current feature information, and query the current description The current feature information corresponding to the information.
  • the recommendation feature generating unit 404 includes: a feature vector generating subunit, configured to obtain the historical feature vector and the current feature corresponding to the historical feature information and the current feature information respectively vector; a recommendation vector splicing subunit, configured to input the historical feature vector and the current feature vector to the splicing deep learning neural network layer for splicing, and splicing to generate a recommended feature vector as the recommended feature information; wherein, the dimension of the recommended feature vector The number is based on the sum of the number of dimensions of the historical feature vector and the number of dimensions of the current feature vector.
  • the recommendation information sending unit 405 includes: a recommendation information generating subunit, configured to use a predetermined recommendation feature recognition neural network to identify the recommendation feature information, and iteratively generate a plurality of recommendation information; wherein, the loss function of the recommendation feature recognition neural network is a cross entropy function; the recommendation information determination subunit is configured to, in response to determining that the loss function converges, stop iterative identification, and determine the finally obtained recommendation information as the first recommendation information; a recommendation information sending subunit, configured to send the first recommendation information to the user.
  • This embodiment exists as an apparatus embodiment corresponding to the foregoing method embodiment. For the same content, reference is made to the description of the foregoing method embodiment, which will not be repeated here.
  • the recommendation information is determined in combination with the user's historical description information, the recommendation information is more accurately provided for the user, and the quality of the recommendation information is improved.
  • FIG. 5 it is a block diagram of an electronic device of a method for generating recommendation information according to an embodiment of the present application.
  • Electronic devices are intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers.
  • Electronic devices may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smart phones, wearable devices, and other similar computing devices.
  • the components shown herein, their connections and relationships, and their functions are by way of example only, and are not intended to limit implementations of the application described and/or claimed herein.
  • the electronic device includes: one or more processors 501, a memory 502, and interfaces for connecting various components, including a high-speed interface and a low-speed interface.
  • the various components are interconnected using different buses and may be mounted on a common motherboard or otherwise as desired.
  • the processor may process instructions executed within the electronic device, including instructions stored in or on memory to display graphical information of the GUI on an external input/output device, such as a display device coupled to the interface.
  • multiple processors and/or multiple buses may be used with multiple memories and multiple memories, if desired.
  • multiple electronic devices may be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system).
  • a processor 501 is taken as an example in FIG. 5 .
  • the memory 502 is the non-transitory computer-readable storage medium provided by the present application.
  • the memory stores instructions executable by at least one processor, so that the at least one processor executes the method for generating recommendation information provided by the present application.
  • the non-transitory computer-readable storage medium of the present application stores computer instructions for causing a computer to execute the method for generating recommendation information provided by the present application.
  • the memory 502 can be used to store non-transitory software programs, non-transitory computer-executable programs, and modules, such as program instructions/modules (such as program instructions/modules corresponding to the method for generating recommendation information in the embodiments of the present application).
  • the processor 501 executes various functional applications and data processing of the server by running the non-transitory software programs, instructions and modules stored in the memory 502, that is, implementing the method for generating recommendation information in the above method embodiments.
  • the memory 502 may include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required for at least one function; the storage data area may store data created according to the use of the electronic device that generates the recommendation information, etc. . Additionally, memory 502 may include high-speed random access memory, and may also include non-transitory memory, such as at least one magnetic disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory 502 may optionally include memory located remotely relative to the processor 501, and these remote memories may connect electronic devices that generate recommendation information through a network. Examples of such networks include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.
  • the electronic device for performing the method of generating recommendation information may further include: an input device 503 and an output device 504 .
  • the processor 501 , the memory 502 , the input device 503 and the output device 504 may be connected by a bus or in other ways, and the connection by a bus is taken as an example in FIG. 5 .
  • the input device 503 can receive input numerical or character information, and generate key signal input related to user settings and function control of the electronic device generating the recommendation information, such as a touch screen, a keypad, a mouse, a trackpad, a touchpad, a pointing stick, One or more input devices such as mouse buttons, trackballs, joysticks, etc.
  • the output device 504 may include a display device, auxiliary lighting devices (eg, LEDs), haptic feedback devices (eg, vibration motors), and the like.
  • the display device may include, but is not limited to, a liquid crystal display (LCD), a light emitting diode (LED) display, and a plasma display. In some implementations, the display device may be a touch screen.
  • Various implementations of the systems and techniques described herein can be implemented in digital electronic circuitry, integrated circuit systems, application specific ASICs (application specific integrated circuits), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include being implemented in one or more computer programs executable and/or interpretable on a programmable system including at least one programmable processor that The processor, which can be a special purpose or general-purpose programmable processor, 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. an output device.
  • the processor which can be a special purpose or general-purpose programmable processor, 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. an output device.
  • machine-readable medium and “computer-readable medium” refer to any computer program product, apparatus, and/or apparatus for providing machine instructions and/or data to a programmable processor ( For example, magnetic disks, optical disks, memories, programmable logic devices (PLDs), including machine-readable media that receive machine instructions as machine-readable signals.
  • machine-readable signal refers to any signal used to provide machine instructions and/or data to a programmable processor.
  • the systems and techniques described herein may be implemented on a computer having a display device (eg, a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user ); and a keyboard and pointing device (eg, a mouse or trackball) through which a user can provide input to the computer.
  • a display device eg, a CRT (cathode ray tube) or LCD (liquid crystal display) monitor
  • a keyboard and pointing device eg, a mouse or trackball
  • Other kinds of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (eg, visual feedback, auditory feedback, or tactile feedback); and can be in any form (including acoustic input, voice input, or tactile input) to receive input from the user.
  • the systems and techniques described herein may be implemented on a computing system that includes back-end components (eg, as a data server), or a computing system that includes middleware components (eg, an application server), or a computing system that includes front-end components (eg, as a data server).
  • back-end components eg, as a data server
  • middleware components eg, an application server
  • front-end components eg, as a data server
  • a user's computer having a graphical user interface or web browser through which a user may interact with implementations of the systems and techniques described herein), or including such backend components, middleware components, Or any combination of front-end components in a computing system.
  • the components of the system may be interconnected by any form or medium of digital data communication (eg, a communication network). Examples of communication networks include: Local Area Networks (LANs), Wide Area Networks (WANs), and the Internet.
  • a computer system can include clients and servers.
  • Clients and servers are generally remote from each other and usually interact through a communication network.
  • the relationship of client and server arises by computer programs running on the respective computers and having a client-server relationship to each other.
  • the current description information input by the user and the historical description information of the user are obtained, and the historical description information is input into the historical feature matching model for processing, Obtain historical feature information, query the current feature information corresponding to the current description information, splicing the historical feature information and the current feature information to generate recommended feature information, and then use a predetermined recommendation feature recognition neural network to identify the recommended feature information, and obtain the first feature.
  • a recommendation information is sent, the first recommendation information is sent to the user, the recommendation information is determined in combination with the user's historical description information, the recommendation information is more accurately provided for the user, and the quality of the recommendation information is improved.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Mathematical Physics (AREA)
  • Computational Linguistics (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • General Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biophysics (AREA)
  • Evolutionary Computation (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Software Systems (AREA)
  • Medical Informatics (AREA)
  • Public Health (AREA)
  • Human Computer Interaction (AREA)
  • Pathology (AREA)
  • Epidemiology (AREA)
  • Primary Health Care (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

生成推荐信息的方法、装置、电子设备及计算机可读存储介质，涉及人工智能领域、自然语言处理技术领域、知识图谱技术领域和大数据技术领域。该方法包括：响应于确定用户非首次访问用户后，获取该用户输入的当前描述信息和该用户的历史描述信息（201），将该历史描述信息输入历史特征匹配模型中进行处理，得到历史特征信息（202），查询该当前描述信息对应的当前特征信息（203），拼接该历史特征信息和该当前特征信息，生成推荐特征信息（204），然后采用预先确定的推荐特征识别神经网络识别该推荐特征信息，得到第一推荐信息并发送该第一推荐信息给该用户（205）。上述方法结合用户的历史描述信息来确定推荐信息，更准确的为用户提供推荐信息，提高推荐信息的质量。

Description

生成推荐信息的方法、装置
本专利申请要求于2020年11月9日提交的、申请号为202011239590.5、发明名称为“生成推荐信息的方法、装置”的中国专利申请的优先权，该申请的全文以引用的方式并入本申请中。
技术领域
本申请涉及人工智能领域，具体涉及自然语言处理技术领域和大数据技术领域，尤其涉及生成推荐信息的方法、装置、电子设备及计算机可读存储介质。
背景技术
患者通过互联网医院进行问诊时，分诊系统会与患者进行简单交互，以便提前获取患者的病例信息，然后分诊系统会根据患者的病例信息进行自然语言处理，进行智能分科室，以代替传统的规则匹配，带来了多方面的收益。
现有技术中通常采用语言识别神经网络对用户的当前描述进行识别，根据得到的识别结果与科室标签进行匹配，完成科室分配、推荐。
发明内容
本申请提供了一种生成推荐信息的方法、装置、电子设备以及存储介质。
第一方面，本申请的实施例提供了一种生成推荐信息的方法，包括：响应于确定用户非首次访问用户，获取该用户输入的当前描述信息和该用户的历史描述信息；将该历史描述信息输入历史特征匹配模型中进行处理，得到历史特征信息；查询该当前描述信息对应的当前特征信息；拼接该历史特征信息和该当前特征信息，生成推荐特征信息；采用预先确定的推荐特征识别神经网络识别该推荐特征信息，得到第一推荐信息，并发送该第一推荐信息给该用户。
在一些实施例中，还包括：响应于该用户为首次访问用户，获取该用 户输入的当前描述信息；查询该当前描述信息对应的当前特征信息；采用该推荐特征识别神经网络识别该当前特征信息，得到第二推荐信息，并发送该第二推荐信息给该用户。
在一些实施例中，将历史描述信息输入历史特征匹配模型中进行处理，得到历史特征信息包括：获取该用户的历史描述信息中的用户描述信息和推荐结果描述信息；将该用户描述信息进行归一化后，输入至构成该历史特征匹配模型的第一深度学习神经网络层进行处理，生成第一特征信息；将该推荐结果描述信息输入至构成该历史特征匹配模型的多层双向变换器编码器层，生成第二特征信息；将该第一特征信息和该第二特征信息输入至构成该历史特征匹配模型的第二深度学习网络层进行拼接，得到该历史特征信息。
在一些实施例中，查询当前描述信息对应的当前特征信息包括：将当前描述信息输入至用于输出当前特征信息的多层双向变换器编码器层，查询该当前描述信息对应的当前特征信息。
在一些实施例中，获取历史特征信息和当前特征信息，拼接生成推荐特征信息包括：分别获取该历史特征信息和该当前特征信息对应的历史特征向量和当前特征向量；将该历史特征向量和该当前特征向量输入至拼接深度学习神经网络层进行拼接，拼接生成推荐特征向量作为推荐特征信息；其中，该推荐特征向量的维度数量基于该历史特征向量的维度数量与该当前特征向量维度数量的加和得到。
在一些实施例中，采用预先确定的推荐特征识别神经网络识别推荐特征信息，得到第一推荐信息，并发送该第一推荐信息给该用户包括：采用预先确定的推荐特征识别神经网络识别该推荐特征信息，迭代生成多个推荐信息；其中，该推荐特征识别神经网络的损失函数为交叉熵函数；响应于确定该损失函数收敛，停止迭代识别，并将最终得到的推荐信息确定为该第一推荐信息，发送该第一推荐信息给该用户。
第二方面，本申请的实施例提供了一种生成推荐信息的装置，包括：描述信息获取单元，被配置成响应于确定用户非首次访问用户，获取该用户输入的当前描述信息和该用户的历史描述信息；历史特征匹配单元，被配置成将该历史描述信息输入历史特征匹配模型中进行处理，得到历史特 征信息；当前特征查询单元，被配置成查询该当前描述信息对应的当前特征信息；推荐特征生成单元，被配置成拼接该历史特征信息和该当前特征信息，生成推荐特征信息；推荐信息发送单元，被配置成采用预先确定的推荐特征识别神经网络识别该推荐特征信息，得到第一推荐信息，并发送该第一推荐信息给该用户。
在一些实施例中，该描述信息获取单元进一步被配置成，响应于该用户为首次访问用户，获取该用户输入的当前描述信息包括；该推荐特征生成单元进一步被配置成，查询该当前描述信息对应的当前特征信息；推荐信息发送单元进一步被配置成，采用该推荐特征识别神经网络识别该当前特征信息，得到第二推荐信息，并发送该第二推荐信息给该用户。
在一些实施例中，该历史特征匹配单元包括：历史信息获取子单元，被配置成获取该用户的历史描述信息中的用户描述信息和推荐结果描述信息；第一特征信息生成子单元，被配置成将该用户描述信息进行归一化后，输入至构成该历史特征匹配模型的第一深度学习神经网络层进行处理，生成第一特征信息；第二特征信息生成子单元，被配置成将该推荐结果描述信息输入至构成该历史特征匹配模型的多层双向变换器编码器层，生成第二特征信息；历史特征信息生成子单元，被配置成将该第一特征信息和该第二特征信息输入至构成该历史特征匹配模型的第二深度学习网络层进行拼接，得到该历史特征信息。
在一些实施例中，该当前特征查询单元进一步被配置成：将当前描述信息输入至用于输出当前特征信息的多层双向变换器编码器层，查询该当前描述信息对应的当前特征信息。
在一些实施例中，该推荐特征生成单元包括：特征向量生成子单元，被配置成分别获取该历史特征信息和该当前特征信息对应的历史特征向量和当前特征向量；推荐向量拼接子单元，被配置成将该历史特征向量和该当前特征向量输入至拼接深度学习神经网络层进行拼接，拼接生成推荐特征向量作为推荐特征信息；其中，该推荐特征向量的维度数量基于该历史特征向量的维度数量与该当前特征向量维度数量的加和得到。
在一些实施例中，该推荐信息发送单元包括：推荐信息生成子单元，被配置成采用预先确定的推荐特征识别神经网络识别该推荐特征信息，迭 代生成多个推荐信息；其中，该推荐特征识别神经网络的损失函数为交叉熵函数；推荐信息确定子单元，被配置成响应于确定该损失函数收敛，停止迭代识别，并将最终得到的推荐信息确定为该第一推荐信息；推荐信息发送子单元，被配置成发送该第一推荐信息给该用户。
第三方面，本申请的实施例提供了一种电子设备，包括：至少一个处理器；以及与上述至少一个处理器通信连接的存储器；其中，该存储器存储有可被上述至少一个处理器执行的指令，该指令被上述至少一个处理器执行，以使上述至少一个处理器能够执行任一实现方式描述的生成推荐信息的方法。
第四方面，本申请的实施例提供了一种存储有计算机指令的非瞬时计算机可读存储介质，包括：该计算机指令用于使该计算机执行任一实现方式描述的生成推荐信息的方法。
本申请在响应于确定用户非首次访问用户后，获取该用户输入的当前描述信息和该用户的历史描述信息，将该历史描述信息输入历史特征匹配模型中进行处理，得到历史特征信息，查询该当前描述信息对应的当前特征信息，拼接该历史特征信息和该当前特征信息，生成推荐特征信息，然后采用预先确定的推荐特征识别神经网络识别该推荐特征信息，得到第一推荐信息，并发送该第一推荐信息给该用户，结合用户的历史描述信息来确定推荐信息，更准确的为用户提供推荐信息，提高推荐信息的质量。
应当理解，本部分所描述的内容并非旨在标识本申请的实施例的关键或重要特征，也不用于限制本申请的范围。本申请的其它特征将通过以下的说明书而变得容易理解。
附图说明
附图用于更好地理解本方案，不构成对本申请的限定。其中：
图1是本申请的实施例可以应用于其中的示例性系统架构；
图2是根据本申请的生成推荐信息的方法的一个实施例的流程图；
图3是根据本申请的生成推荐信息的方法中得到历史特征信息的一个实现方式的流程图；
图4是根据本申请的生成推荐信息的装置的一个实施例的结构示意图；
图5是适于用来实现本申请实施例的生成推荐信息的方法的电子设备的框图。
具体实施方式
以下结合附图对本申请的示范性实施例做出说明，其中包括本申请实施例的各种细节以助于理解，应当将它们认为仅仅是示范性的。因此，本领域普通技术人员应当认识到，可以对这里描述的实施例做出各种改变和修改，而不会背离本申请的范围和精神。同样，为了清楚和简明，以下的描述中省略了对公知功能和结构的描述。
需要说明的是，在不冲突的情况下，本申请中的实施例及实施例中的特征可以相互组合。下面将参考附图并结合实施例来详细说明本申请。
图1示出了可以应用本申请的生成推荐信息的方法、装置、电子设备及计算机可读存储介质的实施例的示例性系统架构100。
如图1所示，系统架构100可以包括终端设备101、102、103，网络104和服务器105。网络104用以在终端设备101、102、103和服务器105之间提供通信链路的介质。网络104可以包括各种连接类型，例如有线、无线通信链路或者光纤电缆等等。
用户可以使用终端设备101、102、103通过网络104与服务器105交互，以实现发送用户的输入信息等目的。终端设备101、102、103上可以安装有与接收调度指示、推荐信息相关的应用，例如服务预约类应用、场景查询类应用、分流指引类应用等。
终端设备101、102、103可以是硬件，也可以是软件。硬件时，可以是具有显示屏的各种电子设备，包括但不限于智能手机、平板电脑、膝上型便携计算机和台式计算机等等。当终端设备101、102、103为软件时，可以安装在上述所列举的电子设备中。其可以实现成多个软件或软件模块(例如发送用户的输入信息等)，也可以实现成单个软件或软件模块。在此不做具体限定。
服务器105可以是提供各种服务的服务器，例如为终端设备101、102、103提供推送信息的服务器。例如响应于确定用户非首次访问用户后，获取该用户输入的当前描述信息和该用户的历史描述信息，将该历史描述信 息输入历史特征匹配模型中进行处理，得到历史特征信息，查询该当前描述信息对应的当前特征信息，拼接该历史特征信息和该当前特征信息，生成推荐特征信息；采用预先确定的推荐特征识别神经网络识别该推荐特征信息，得到第一推荐信息并发送该第一推荐信息给该用户。
需要说明的是，本申请的实施例所提供的生成推荐信息的方法一般由服务器105执行，相应地，生成推荐信息的装置一般设置于服务器105中。
需要说明的是，服务器可以是硬件，也可以是软件。当服务器为硬件时，可以实现成多个服务器组成的分布式服务器集群，也可以实现成单个服务器。当服务器为软件时，可以实现成例如用来提供分布式服务的多个软件或软件模块，也可以实现成单个软件或软件模块。在此不做具体限定。
此外，生成推荐信息的方法也可以由终端设备101、102、103执行，相应地，生成推荐信息的装置也可以设置于终端设备101、102、103中。此时，示例性系统架构100也可以不包括服务器105和网络104。
应该理解，图1中的终端设备、网络和服务器的数目仅仅是示意性的。根据实现需要，可以具有任意数目的终端设备、网络和服务器。
继续参考图2，其示出了根据本申请的生成推荐信息的方法的一个实施例流程200。该生成推荐信息的方法，包括以下步骤：
步骤201，响应于确定用户非首次访问用户，获取该用户输入的当前描述信息和该用户的历史描述信息。
在本实施例中，生成推荐信息的方法的执行主体(例如图1所示的服务器105)会首先判断接收推送信息的用户是否为首次访问的用户，通常可以根据上述执行主体的本地中是否储存有该用户的历史访问记录的方式进行判断，也可以以人机交互的方式来向该用户进行求证，以辅助上述执行主体的判断。
在确定该用户为非首次访问的用户时，获取该用户输入的当前描述信息，当前描述信息指的是本次由用户提供的、以希望获得对应匹配的推荐信息的描述信息，通常当前描述信息是用户通过使用的人机交互设备(例如图1所示的终端设备101、102、103)进行输入，然后与上述执行主体进行通信(例如通过网络104)完成当前描述信息的发送的，以便于用户可以通过使用的人机交互设备完成推荐信息的获取工作。
示例性的，在以推荐信息为科室的分类、导诊信息时，当前描述信息可以为该用户输入的本次希望诊断的疾病信息、症状信息等，在以推荐信息为餐饮推荐信息时，当前描述信息可以为用户希望品尝的菜品口味、忌口信息等信息。
在获取该用户输入的当前描述信息后，可以同时或依次获取用户的历史描述信息，历史描述信息为相对于该用户本次的描述信息存在的描述信息，同样以推荐信息为科室的分类、导诊信息时，历史描述信息可以为用户的历史科室分配结果情况、症状情况、病例信息和诊断结果等，应当理解的是，历史描述信息通常为保存在上述执行主体的本地的描述信息，实现上述执行主体对用户的信息进行收集、储存，以便于后续使用，同样的历史描述信息也可以由用户进行输入或者保存在用户使用的人机交互设备本地，上述执行主体可以以通信交互的方式从该人机交互设备获取对应的历史描述信息，本申请对此不做限定。
步骤202，将历史描述信息输入历史特征匹配模型中进行处理，得到历史特征信息。
在本实施例中，在获取到用户的历史描述信息后，将该历史描述信息输入历史特征匹配模型中进行处理，历史特征匹配模型通常为将大量不同用户的历史进行收集并储存的模型，在输入全部或者部分的用户的历史描述信息后，该历史特征匹配模型可以获取更全面的该用户的历史描述信息，并进行对应的连接、压缩，生成一个尽可能多的包含有用户的历史信息的历史特征信息，以便于后续采用该历史特征信息来表征用户的历史描述信息、历史推荐信息的情况，完成基于历史数据的用户的行为分析。
应当理解的是，在一般的应用场景下，用户的历史描述信息和历史推荐信息生成结果都保存在上述执行主体中，上述执行主体可以对应的生成一个历史数据库对这些内容进行记载，以便于生成该历史特征匹配模型。
步骤203，查询当前描述信息对应的当前特征信息。
在本实施例中，在获取到用户的当前描述信息后，可以采用特征提取、特征识别神经网络等方式对当前描述信息进行对应的特征提取，以得到表征该用户本次期望被推荐的行为的当前特征信息，以科室分类、导诊应用场景为例，用户的当前描述信息可以为本次希望诊断、解决的具体症状信 息、用户主观期望前往的科室信息和用户对于科室选择的特殊要求信息等信息。
步骤204，拼接该历史特征信息和该当前特征信息，生成推荐特征信息。
在本实施例中，在上述步骤202和步骤203中分别得到历史特征信息和当前特征信息后，可以结合特征信息的表现形式选择合适的拼接形式对历史特征信息和当前特征信息进行拼接，例如历史特征信息和当前特征信息的表现形式都为信息集合时，可以将量集合进行合并以得到推荐特征信息集合，或者历史特征信息和当前特征信息的表现形式都为向量形式时，可以基于历史特征信息和当前特征信息的维度数量信息进行叠加拼接(例如历史特征信息的维度数量为A，当前特征信息的维度数量为B，则拼接得到的推荐特征信息的维度数量为C＝A+B)或者覆盖拼接(例如历史特征信息的维度数量为A，当前特征信息的维度数量同样为A，则将历史特征信息和当前特征信息对应维度的内容进行叠加)的方式实现历史特征信息和当前特征信息的拼接，以得到推荐特征信息。
其中，在根据历史特征信息和当前特征信息的形式不同，还可以预先确定不同的约束规则，以调整得到的推荐信息的形式，例如在历史特征信息和当前特征信息的表现形式都为向量形式时，可以预先确定推荐信息的形式为预设维度数量的特征向量，此时，在获取到历史特征信息和当前特征信息的拼接结果后，采用压缩或者放大等形式对拼接结果进行处理，以得到表现形式为预设维度数量的特征向量作为推荐信息。
步骤205，采用预先确定的推荐特征识别神经网络识别该推荐特征信息，得到第一推荐信息并发送该第一推荐信息给该用户。
在本实施例中，预先确定的推荐特征识别神经网络为预先根据得到的推荐特征信息的形式确定的推荐特征神经网络，预先可以将历史数据中的推荐特征信息作为输入，以得到的第一推荐信息作为输出对该推荐特征神经网络进行训练，以实现在得到上述步骤204中得到推荐特征信息后，可以使用该推荐特征识别神经网络对该推荐特征信息进行识别，得到最终的推荐信息。
在得到第一推荐信息后，发送该第一推荐信息给该用户以完成推荐信 息的生成工作，为该用户提供推荐信息。
本申请实施例提供的生成推荐信息的方法，在响应于确定用户非首次访问用户后，获取该用户输入的当前描述信息和该用户的历史描述信息，将该历史描述信息输入历史特征匹配模型中进行处理，得到历史特征信息，查询该当前描述信息对应的当前特征信息，拼接该历史特征信息和该当前特征信息，生成推荐特征信息，然后采用预先确定的推荐特征识别神经网络识别该推荐特征信息，得到第一推荐信息，并发送该第一推荐信息给该用户，结合用户的历史描述信息来确定推荐信息，更准确的为用户提供推荐信息，提高推荐信息的质量。
在本实施例的一些可选实现方式中，还包括：响应于该用户为首次访问用户，获取该用户输入的当前描述信息；查询该当前描述信息对应的当前特征信息；采用该推荐特征识别神经网络识别该当前特征信息，得到第二推荐信息，并发送该第二推荐信息给该用户。
具体的，在上述步骤201中确定到该用户为首先访问用户时，因不存在该用户的历史描述信息，则直接获取用户输入的当前描述进行，同样采用上述步骤203中的方法来获取该用户对应的当前特征信息，因推荐特征信息为当前特征信息和历史特征信息拼接而来的，在不存在历史特征信息的情况下，则相当于当前特征信息中的内容与推荐特征信息的内容相同，同样采用上述步骤205中的推荐特征识别神经网络对该推荐特征进行识别，以获取第二推荐信息，并发送该第二推荐信息给用户以完成信息推送工作，以确保在不存在历史描述信息时，即用户为首次访问用户时的推荐工作，避免因缺少历史数据造成的无法生成推荐信息的问题。
在本实施例的一些可选实现方式中，参考图3，其中示出了一种采用历史特征匹配模型对历史描述信息进行处理，以得到历史特征信息的流程300，具体包括：
步骤301，获取用户的历史描述信息中的用户描述信息和推荐结果描述信息。
具体的，用户的历史描述信息中通常具有大量的信息内容，因此为了方便识别，可以将这些信息分为两类，即用户描述信息和推荐结果描述信息，以推荐信息为科室分类、导诊的应用场景为例，用户描述信息为用户 的性别、年龄、距离最近一次访问天数、最近一次访问的科室、最近一次问诊的医生所属科室、对最近一次问诊的医生的评价、是否购药开方等信息，推荐结果描述信息可以为用户最近一次的病例描述信息、症状信息、诊断信息以开具药品明细信息等。
步骤302，将该用户描述信息进行归一化后输入至构成该历史特征匹配模型的第一深度学习神经网络层进行处理，生成第一特征信息。
具体的，因用户描述信息中记载有比较明确、标准可识别的实体内容较少，因此将上述步骤301中获取到的用户描述信息进行例如最大最小归一化、语义归一化的处理方式进行归一化处理以得到标准的可被第一深度学习神经网络进行识别、提取的处理结果，深度学习神经网络(Deep Neural Networks，简称DNN)，然后将该结果输入到构成历史特征匹配模型的第一深度学习神经网络中，深度学习神经网络(Deep Neural Networks，简称DNN)可以实现对该处理结果进行识别、提取，以得到用以表征用户描述信息的第一特征信息。
示例性的，基于患者的性别、年龄、距离最近一次访问天数、最近一次访问的科室(二进制编码，4位表示)、最近一次问诊的医生所属科室(二进制编码，4位表示)、对最近一次问诊的医生的评价、是否购药开方等特征，然后进行最大最小归一化，得到的归一化处理结果总共有15个维度，输入DNN层(其中包含有两个隐藏层：第一个隐藏层60个神经元，该层使用线性整流激活函数(Rectified Linear Unit，简称Relu)；第二个隐藏层30个神经元，激活函数为Relu；输出层的维度数量为5)，输出层输出维度数量为5的第一特征信息。
步骤303，将该推荐结果描述信息输入至构成该历史特征匹配模型的多层双向变换器编码器层，生成第二特征信息。
具体的，因推荐结果描述信息中记载的内容通常为标准、准确的结果内容，因此将推荐结果描述信息输入至构成该历史特征匹配模型的多层双向变换器编码器层(Bidirectional Encoder Representations from Transformers，简称Bert)进行处理，以得到用以表征推荐结果描述信息的第二特征信息。
示例性的，将用户最近一次病例描述、症状和药品明细进行拼接，作 为Bert层的输入，目的学习用户上次的病例特征，根据实际情况分析，用户在短期内病情存在一定的相关性。Bert层的最终输出维度数量为10的第二特征信息。
步骤304，将该第一特征信息和该第二特征信息输入至构成该历史特征匹配模型的第二深度学习网络层进行拼接，得到该历史特征信息。
具体的，将上述步骤302和步骤303中得到的第一特征信息和第二特征信息输入至第二深度学习网络层中进行拼接，以得到最终的历史特征信息，该历史特征信息中包括有用以表征用户描述信息的第一特征信息，和用以表征推荐结果描述信息的第二特征信息的相关内容。
示例性的，将上述DNN和Bert模型的输出拼接，即将第一特征信息和第二特征信息进行拼接，一方面进行联合训练学习，抽象提取两部分模型的组合特征；另一方面将上侧两部分特征进行压缩，输出维度数量为5的历史特征信息，以实现减少特征维度，提高运行性能的目的。该部分DNN模型共三个隐藏层：第一个隐藏层200个神经元，该层使用Relu激活函数；第二个隐藏层100个神经元，同样使用Relu激活函数；第三个隐藏层60个神经元，同样使用Relu激活函数；输出维度数量为5的历史特征信息，作为用户的历史特征信息。
在本实现方式中，通过历史特征匹配模型基于用户描述信息和推荐结果信息来确定该用户历史描述信息的历史特征信息，以便于后续采用该历史特征信息来更全面的反映出用户的情况，结合用户的历史数据来确定推荐信息，以解决现有技术中因用户的录入信息过少，无法生成高质量的推荐信息、准确为用户推荐内容的问题。
在本实施例的一些可选实现方式中，查询当前描述信息对应的当前特征信息包括：将当前描述信息输入至用于输出当前特征信息的多层双向变换器编码器层，查询该当前描述信息对应的当前特征信息。
具体的，在获取到用户输入的当前描述信息后，将推荐结果描述信息输入用于输出当前特征信息的另一个Bert中，进行处理，以得到用以表征当前描述信息对应的当前特征信息，以实现将当前描述信息处理为与历史特征信息近似的表现形式的当前特征信息，便于将历史特征信息和当前特征信息进行拼接，生成推荐特征信息。
在本实施例的一些可选实现方式中，获取该历史特征信息和该当前特征信息，拼接生成推荐特征信息包括：分别获取该历史特征信息和该当前特征信息对应的历史特征向量和当前特征向量；将该历史特征向量和该当前特征向量输入至拼接深度学习神经网络层进行拼接，拼接生成推荐特征向量作为推荐特征信息；其中，该推荐特征向量的维度数量基于该历史特征向量的维度数量与该当前特征向量维度数量的加和得到。
具体的，可以确定以向量作为历史特征信息和当前特征信息的表现形式，然后分别获取该历史特征信息和该当前特征信息对应的历史特征向量和当前特征向量，然后将历史特征向量和当前特征向量直接进行拼接，即得到的作为推荐特征信息的推荐特征向量的维度数量为历史特征向量的维度数量和当前特征向量维度数量的加和，通过这种直接拼接的方式，高效率的生成了包含有全部历史特征向量和当前特征向量信息的推荐特征向量，平衡了推荐特征信息的生成效率和质量之间的关系。
示例性的，采用DNN层将得到的以5维向量表示的第一特征信息和以10维向量表示的第二特征信息进行拼接，得到以15维向量表示的特征信息。
在本实施例的一些可选实现方式中，采用预先确定的推荐特征识别神经网络识别该推荐特征信息，得到第一推荐信息，并发送该第一推荐信息给该用户包括：采用预先确定的推荐特征识别神经网络识别该推荐特征信息，迭代生成多个推荐信息；其中，该推荐特征识别神经网络的损失函数为交叉熵函数；响应于确定该损失函数收敛，停止迭代识别，并将最终得到的推荐信息确定为该第一推荐信息，发送该第一推荐信息给该用户。
具体的，在获取到推荐特征信息后，以迭代识别的方式采用预先确定的以交叉熵函数为损失函数的推荐特征识别网络对该推荐特征进行迭代识别，响应于确定该损失函数收敛，停止迭代识别，将最终得到的推荐信息作为第一推荐信息，发送给用户，以迭代识别的方式提升推荐特征识别网络输出的第一推荐信息的质量。
为加深理解，本申请还结合一个具体应用场景，给出了一种具体的实现方案。在该具体应用场景下，非首次进行访问的用户u1通过自身使用的人机交互设备输入的当前描述信息“昨夜到今晨连续拉肚子”，并发送给 生成推荐信息的方法的执行主体(简称推荐执行主体)。
推荐主体确定该用户u1为非首次访问用户，获取用户u1发送的当前描述信息“昨夜到今晨连续拉肚子”以及从本地获取到历史描述信息为“性别男，曾经就诊过内科科室，最近一次来的症状描述信息为连续腹泻、胃阵痛，诊断结果为浅表性胃炎并开具胃药”。
分别将历史描述信息中的用户描述信息“性别男，曾经就诊过内科科室”以及推荐结果描述信息“最近一次来的症状描述信息为连续腹泻、胃阵痛，诊断结果为浅表性胃炎并开具胃药”输入至构成历史特征匹配模型的第一深度学习神经网络层和构成历史特征匹配模型的多层双向变换器编码器层，得到第一特征信息、第二特征信息。
将第一特征信息、第二特征信息输入至构成该历史特征匹配模型的第二深度学习网络层进行拼接，得到维度数量为5的历史特征向量作为历史特征信息。
将当前描述信息“昨夜到今晨连续拉肚子”，输入至用于输出当前特征信息的多层双向变换器编码器层，查询该当前描述信息对应的以维度数量为10的当前特征向量表示的当前特征信息。
然后拼接该历史特征信息和该当前特征信息，生成以维度数量为10的推荐特征向量进行表示的推荐特征信息，并采用预先确定的推荐特征识别神经网络识别该推荐特征信息，得到第一推荐信息，并发送该第一推荐信息给该用户。
通过本应用场景可以看出，本申请实施例提供的生成推荐信息的方法，该用户输入的当前描述信息和该用户的历史描述信息，将该历史描述信息输入历史特征匹配模型中进行处理，得到历史特征信息，查询该当前描述信息对应的当前特征信息，拼接该历史特征信息和该当前特征信息，生成推荐特征信息，然后采用预先确定的推荐特征识别神经网络识别该推荐特征信息，得到第一推荐信息，并发送该第一推荐信息给该用户，结合用户的历史描述信息来确定推荐信息，更准确的为用户提供推荐信息，提高推荐信息的质量。
如图4所示，本实施例的生成推荐信息的装置400可以包括：描述信息获取单元401，被配置成响应于确定用户非首次访问用户，获取该用户 输入的当前描述信息和该用户的历史描述信息；历史特征匹配单元402，被配置成将该历史描述信息输入历史特征匹配模型中进行处理，得到历史特征信息；当前特征查询单元403，被配置成查询该当前描述信息对应的当前特征信息；推荐特征生成单元404，被配置成拼接该历史特征信息和该当前特征信息，生成推荐特征信息；推荐信息发送单元405，被配置成采用预先确定的推荐特征识别神经网络识别该推荐特征信息，得到第一推荐信息并发送该第一推荐信息给该用户。
在本实施例的一些可选的实现方式中，上述生成推荐信息的装置还包括：该描述信息获取单元401进一步被配置成，响应于该用户为首次访问用户，获取该用户输入的当前描述信息包括；该推荐特征生成单元404进一步被配置成，查询该当前描述信息对应的当前特征信息；推荐信息发送单元405进一步被配置成，采用该推荐特征识别神经网络识别该当前特征信息，得到第二推荐信息，并发送该第二推荐信息给该用户。
在本实施例的一些可选的实现方式中，历史特征匹配单元包括：历史信息获取子单元，被配置成获取该用户的历史描述信息中的用户描述信息和推荐结果描述信息；第一特征信息生成子单元，被配置成将将该用户描述信息进行归一化后，输入至构成该历史特征匹配模型的第一深度学习神经网络层进行处理，生成第一特征信息；第二特征信息生成子单元，被配置成将该推荐结果描述信息输入至构成该历史特征匹配模型的多层双向变换器编码器层，生成第二特征信息；历史特征信息生成子单元，被配置成将该第一特征信息和该第二特征信息输入至构成该历史特征匹配模型的第二深度学习网络层进行拼接，得到该历史特征信息。
在本实施例的一些可选的实现方式中，推送信息发送单元进一步被配置成：采用概率图模型对该医学状态实体进行排序，根据排序结果选取预设数量的该医学状态实体生成推送信息集合；发送该推送集合给该用户。
在本实施例的一些可选的实现方式中，该当前特征查询单元403进一步被配置成：将当前描述信息输入至用于输出当前特征信息的多层双向变换器编码器层，查询该当前描述信息对应的当前特征信息。
在本实施例的一些可选的实现方式中，该推荐特征生成单元404中包括：特征向量生成子单元，被配置成分别获取该历史特征信息和该当前特 征信息对应的历史特征向量和当前特征向量；推荐向量拼接子单元，被配置成将该历史特征向量和该当前特征向量输入至拼接深度学习神经网络层进行拼接，拼接生成推荐特征向量作为推荐特征信息；其中，该推荐特征向量的维度数量基于该历史特征向量的维度数量与该当前特征向量维度数量的加和得到。
在本实施例的一些可选的实现方式中，该推荐信息发送单元405中包括：推荐信息生成子单元，被配置成采用预先确定的推荐特征识别神经网络识别该推荐特征信息，迭代生成多个推荐信息；其中，该推荐特征识别神经网络的损失函数为交叉熵函数；推荐信息确定子单元，被配置成响应于确定该损失函数收敛，停止迭代识别，并将最终得到的推荐信息确定为该第一推荐信息；推荐信息发送子单元，被配置成发送该第一推荐信息给该用户。
本实施例作为对应于上述方法实施例的装置实施例存在，相同内容参考对于上述方法实施例的说明，对此不再赘述。通过本申请实施例提供的生成推荐信息的装置，结合用户的历史描述信息来确定推荐信息，更准确的为用户提供推荐信息，提高推荐信息的质量。
如图5所示，是根据本申请实施例的生成推荐信息的方法的电子设备的框图。电子设备旨在表示各种形式的数字计算机，诸如，膝上型计算机、台式计算机、工作台、个人数字助理、服务器、刀片式服务器、大型计算机、和其它适合的计算机。电子设备还可以表示各种形式的移动装置，诸如，个人数字处理、蜂窝电话、智能电话、可穿戴设备和其它类似的计算装置。本文所示的部件、它们的连接和关系、以及它们的功能仅仅作为示例，并且不意在限制本文中描述的和/或者要求的本申请的实现。
如图5所示，该电子设备包括：一个或多个处理器501、存储器502，以及用于连接各部件的接口，包括高速接口和低速接口。各个部件利用不同的总线互相连接，并且可以被安装在公共主板上或者根据需要以其它方式安装。处理器可以对在电子设备内执行的指令进行处理，包括存储在存储器中或者存储器上以在外部输入/输出装置(诸如，耦合至接口的显示设备)上显示GUI的图形信息的指令。在其它实施方式中，若需要，可以将多个处理器和/或多条总线与多个存储器和多个存储器一起使用。同样，可 以连接多个电子设备，各个设备提供部分必要的操作(例如，作为服务器阵列、一组刀片式服务器、或者多处理器系统)。图5中以一个处理器501为例。
存储器502即为本申请所提供的非瞬时计算机可读存储介质。其中，该存储器存储有可由至少一个处理器执行的指令，以使上述至少一个处理器执行本申请所提供的生成推荐信息的方法。本申请的非瞬时计算机可读存储介质存储计算机指令，该计算机指令用于使计算机执行本申请所提供的生成推荐信息的方法。
存储器502作为一种非瞬时计算机可读存储介质，可用于存储非瞬时软件程序、非瞬时计算机可执行程序以及模块，如本申请实施例中的生成推荐信息的方法对应的程序指令/模块(例如，图4所示的描述信息获取单元401、历史特征匹配单元402、当前特征查询单元403、推荐特征生成单元404和推荐信息发送单元405)。处理器501通过运行存储在存储器502中的非瞬时软件程序、指令以及模块，从而执行服务器的各种功能应用以及数据处理，即实现上述方法实施例中的生成推荐信息的方法。
存储器502可以包括存储程序区和存储数据区，其中，存储程序区可存储操作系统、至少一个功能所需要的应用程序；存储数据区可存储根据生成推荐信息的电子设备的使用所创建的数据等。此外，存储器502可以包括高速随机存取存储器，还可以包括非瞬时存储器，例如至少一个磁盘存储器件、闪存器件、或其他非瞬时固态存储器件。在一些实施例中，存储器502可选包括相对于处理器501远程设置的存储器，这些远程存储器可以通过网络连接生成推荐信息的电子设备。上述网络的实例包括但不限于互联网、企业内部网、局域网、移动通信网及其组合。
用于执行生成推荐信息的方法的电子设备还可以包括：输入装置503和输出装置504。处理器501、存储器502、输入装置503和输出装置504可以通过总线或者其他方式连接，图5中以通过总线连接为例。
输入装置503可接收输入的数字或字符信息，以及产生与生成推荐信息的电子设备的用户设置以及功能控制有关的键信号输入，例如触摸屏、小键盘、鼠标、轨迹板、触摸板、指示杆、一个或者多个鼠标按钮、轨迹 球、操纵杆等输入装置。输出装置504可以包括显示设备、辅助照明装置(例如，LED)和触觉反馈装置(例如，振动电机)等。该显示设备可以包括但不限于，液晶显示器(LCD)、发光二极管(LED)显示器和等离子体显示器。在一些实施方式中，显示设备可以是触摸屏。
此处描述的系统和技术的各种实施方式可以在数字电子电路系统、集成电路系统、专用ASIC(专用集成电路)、计算机硬件、固件、软件、和/或它们的组合中实现。这些各种实施方式可以包括：实施在一个或者多个计算机程序中，该一个或者多个计算机程序可在包括至少一个可编程处理器的可编程系统上执行和/或解释，该可编程处理器可以是专用或者通用可编程处理器，可以从存储系统、至少一个输入装置、和至少一个输出装置接收数据和指令，并且将数据和指令传输至上述存储系统、上述至少一个输入装置、和上述至少一个输出装置。
这些计算程序(也称作程序、软件、软件应用、或者代码)包括可编程处理器的机器指令，并且可以利用高级过程和/或面向对象的编程语言、和/或汇编/机器语言来实施这些计算程序。如本文使用的，术语“机器可读介质”和“计算机可读介质”指的是用于将机器指令和/或数据提供给可编程处理器的任何计算机程序产品、设备、和/或装置(例如，磁盘、光盘、存储器、可编程逻辑装置(PLD))，包括，接收作为机器可读信号的机器指令的机器可读介质。术语“机器可读信号”指的是用于将机器指令和/或数据提供给可编程处理器的任何信号。
为了提供与用户的交互，可以在计算机上实施此处描述的系统和技术，该计算机具有：用于向用户显示信息的显示装置(例如，CRT(阴极射线管)或者LCD(液晶显示器)监视器)；以及键盘和指向装置(例如，鼠标或者轨迹球)，用户可以通过该键盘和该指向装置来将输入提供给计算机。其它种类的装置还可以用于提供与用户的交互；例如，提供给用户的反馈可以是任何形式的传感反馈(例如，视觉反馈、听觉反馈、或者触觉反馈)；并且可以用任何形式(包括声输入、语音输入或者、触觉输入)来接收来自用户的输入。
可以将此处描述的系统和技术实施在包括后台部件的计算系统(例如， 作为数据服务器)、或者包括中间件部件的计算系统(例如，应用服务器)、或者包括前端部件的计算系统(例如，具有图形用户界面或者网络浏览器的用户计算机，用户可以通过该图形用户界面或者该网络浏览器来与此处描述的系统和技术的实施方式交互)、或者包括这种后台部件、中间件部件、或者前端部件的任何组合的计算系统中。可以通过任何形式或者介质的数字数据通信(例如，通信网络)来将系统的部件相互连接。通信网络的示例包括：局域网(LAN)、广域网(WAN)和互联网。
计算机系统可以包括客户端和服务器。客户端和服务器一般远离彼此并且通常通过通信网络进行交互。通过在相应的计算机上运行并且彼此具有客户端-服务器关系的计算机程序来产生客户端和服务器的关系。
根据本申请实施例的技术方案，在响应于确定用户非首次访问用户后，获取该用户输入的当前描述信息和该用户的历史描述信息，将该历史描述信息输入历史特征匹配模型中进行处理，得到历史特征信息，查询该当前描述信息对应的当前特征信息，拼接该历史特征信息和该当前特征信息，生成推荐特征信息，然后采用预先确定的推荐特征识别神经网络识别该推荐特征信息，得到第一推荐信息，并发送该第一推荐信息给该用户，结合用户的历史描述信息来确定推荐信息，更准确的为用户提供推荐信息，提高推荐信息的质量。
应该理解，可以使用上面所示的各种形式的流程，重新排序、增加或删除步骤。例如，本申请中记载的各步骤可以并行地执行也可以顺序地执行也可以不同的次序执行，只要能够实现本申请公开的技术方案所期望的结果，本文在此不进行限制。
上述具体实施方式，并不构成对本申请保护范围的限制。本领域技术人员应该明白的是，根据设计要求和其他因素，可以进行各种修改、组合、子组合和替代。任何在本申请的精神和原则之内所作的修改、等同替换和改进等，均应包含在本申请保护范围之内。

Claims (14)

  1. 一种生成推荐信息的方法，包括：
    响应于确定用户非首次访问用户，获取所述用户输入的当前描述信息和所述用户的历史描述信息；
    将所述历史描述信息输入历史特征匹配模型中进行处理，得到历史特征信息；
    查询所述当前描述信息对应的当前特征信息；
    拼接所述历史特征信息和所述当前特征信息，生成推荐特征信息；
    采用预先确定的推荐特征识别神经网络识别所述推荐特征信息，得到第一推荐信息，并发送所述第一推荐信息给所述用户。
  2. 根据权利要求1所述的方法，还包括：
    响应于所述用户为首次访问用户，获取所述用户输入的当前描述信息；
    查询所述当前描述信息对应的当前特征信息；
    采用所述推荐特征识别神经网络识别所述当前特征信息，得到第二推荐信息，并发送所述第二推荐信息给所述用户。
  3. 根据权利要求1所述的方法，其中，将所述历史描述信息输入历史特征匹配模型中进行处理，得到历史特征信息包括：
    获取所述用户的历史描述信息中的用户描述信息和推荐结果描述信息；
    将所述用户描述信息进行归一化后，输入至构成所述历史特征匹配模型的第一深度学习神经网络层进行处理，生成第一特征信息；
    将所述推荐结果描述信息输入至构成所述历史特征匹配模型的多层双向变换器编码器层，生成第二特征信息；
    将所述第一特征信息和所述第二特征信息输入至构成所述历史特征匹配模型的第二深度学习网络层进行拼接，得到所述历史特征信息。
  4. 根据权利要求1所述的方法，其中，所述查询所述当前描述信息对应的当前特征信息包括：
    将当前描述信息输入至用于输出当前特征信息的多层双向变换器编码器层，查询所述当前描述信息对应的当前特征信息。
  5. 根据权利要求1所述的方法，其中，所述获取所述历史特征信息和所述当前特征信息，拼接生成推荐特征信息包括：
    分别获取所述历史特征信息和所述当前特征信息对应的历史特征向量和当前特征向量；
    将所述历史特征向量和所述当前特征向量输入至拼接深度学习神经网络层进行拼接，拼接生成推荐特征向量作为推荐特征信息；其中，所述推荐特征向量的维度数量基于所述历史特征向量的维度数量与所述当前特征向量维度数量的加和得到。
  6. 根据权利要求1所述的方法，其中，所述采用预先确定的推荐特征识别神经网络识别所述推荐特征信息，得到第一推荐信息，并发送所述第一推荐信息给所述用户包括：
    采用预先确定的推荐特征识别神经网络识别所述推荐特征信息，迭代生成多个推荐信息；其中，所述推荐特征识别神经网络的损失函数为交叉熵函数；
    响应于确定所述损失函数收敛，停止迭代识别，并将最终得到的推荐信息确定为所述第一推荐信息，发送所述第一推荐信息给所述用户。
  7. 一种生成推荐信息的装置，包括：
    描述信息获取单元，被配置成响应于确定用户非首次访问用户，获取所述用户输入的当前描述信息和所述用户的历史描述信息；
    历史特征匹配单元，被配置成将所述历史描述信息输入历史特征匹配模型中进行处理，得到历史特征信息；
    当前特征查询单元，被配置成查询所述当前描述信息对应的当前 特征信息；
    推荐特征生成单元，被配置成拼接所述历史特征信息和所述当前特征信息，生成推荐特征信息；
    推荐信息发送单元，被配置成采用预先确定的推荐特征识别神经网络识别所述推荐特征信息，得到第一推荐信息并发送所述第一推荐信息给所述用户。
  8. 根据权利要求7所述的装置，还包括：
    所述描述信息获取单元进一步被配置成，响应于所述用户为首次访问用户，获取所述用户输入的当前描述信息包括；
    所述推荐特征生成单元进一步被配置成，查询所述当前描述信息对应的当前特征信息；
    推荐信息发送单元进一步被配置成，采用所述推荐特征识别神经网络识别所述当前特征信息，得到第二推荐信息，并发送所述第二推荐信息给所述用户。
  9. 根据权利要求7所述的装置，其中，所述历史特征匹配单元包括：
    历史信息获取子单元，被配置成获取所述用户的历史描述信息中的用户描述信息和推荐结果描述信息；
    第一特征信息生成子单元，被配置成将将所述用户描述信息进行归一化后，输入至构成所述历史特征匹配模型的第一深度学习神经网络层进行处理，生成第一特征信息；
    第二特征信息生成子单元，被配置成将所述推荐结果描述信息输入至构成所述历史特征匹配模型的多层双向变换器编码器层，生成第二特征信息；
    历史特征信息生成子单元，被配置成将所述第一特征信息和所述第二特征信息输入至构成所述历史特征匹配模型的第二深度学习网络层进行拼接，得到所述历史特征信息。
  10. 根据权利要求7所述的装置，其中，所述当前特征查询单元进一步被配置成：
    将当前描述信息输入至用于输出当前特征信息的多层双向变换器编码器层，查询所述当前描述信息对应的当前特征信息。
  11. 根据权利要求7所述的装置，其中，所述推荐特征生成单元包括：
    特征向量生成子单元，被配置成分别获取所述历史特征信息和所述当前特征信息对应的历史特征向量和当前特征向量；
    推荐向量拼接子单元，被配置成将所述历史特征向量和所述当前特征向量输入至拼接深度学习神经网络层进行拼接，拼接生成推荐特征向量作为推荐特征信息；其中，所述推荐特征向量的维度数量基于所述历史特征向量的维度数量与所述当前特征向量维度数量的加和得到。
  12. 根据权利要求7所述的装置，其中，所述推荐信息发送单元包括：
    推荐信息生成子单元，被配置成采用预先确定的推荐特征识别神经网络识别所述推荐特征信息，迭代生成多个推荐信息；其中，所述推荐特征识别神经网络的损失函数为交叉熵函数；
    推荐信息确定子单元，被配置成响应于确定所述损失函数收敛，停止迭代识别，并将最终得到的推荐信息确定为所述第一推荐信息；
    推荐信息发送子单元，被配置成发送所述第一推荐信息给所述用户。
  13. 一种电子设备，包括：
    至少一个处理器；以及
    与所述至少一个处理器通信连接的存储器；其中，所述存储器存储有可被所述至少一个处理器执行的指令，所述指令被所述至少一个处理器执行，以使所述至少一个处理器能够执行权利要求1-6中任一 项所述的生成推荐信息的方法。
  14. 一种存储有计算机指令的非瞬时计算机可读存储介质，包括：所述计算机指令用于使所述计算机执行权利要求1-6中任一项所述的生成推荐信息的方法。
PCT/CN2021/128399 2020-11-09 2021-11-03 生成推荐信息的方法、装置 WO2022095893A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202011239590.5 2020-11-09
CN202011239590.5A CN112287232B (zh) 2020-11-09 2020-11-09 生成推荐信息的方法、装置

Publications (1)

Publication Number Publication Date
WO2022095893A1 true WO2022095893A1 (zh) 2022-05-12

Family

ID=74351793

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/128399 WO2022095893A1 (zh) 2020-11-09 2021-11-03 生成推荐信息的方法、装置

Country Status (2)

Country Link
CN (1) CN112287232B (zh)
WO (1) WO2022095893A1 (zh)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115274086A (zh) * 2022-09-27 2022-11-01 无码科技（杭州）有限公司 一种智能导诊方法及系统
CN115687674A (zh) * 2022-12-20 2023-02-03 昆明勤砖晟信息科技有限公司 服务于智慧云服务平台的大数据需求分析方法及系统
CN116450925A (zh) * 2022-12-27 2023-07-18 深圳市网新新思软件有限公司 基于人工智能的用户关系分析方法及系统
CN116629810A (zh) * 2023-07-24 2023-08-22 中建五局第三建设有限公司 基于建筑办公系统的操作推荐方法、装置、设备和介质

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112287232B (zh) * 2020-11-09 2024-04-16 北京京东拓先科技有限公司 生成推荐信息的方法、装置
CN113821731A (zh) * 2021-11-23 2021-12-21 湖北亿咖通科技有限公司 信息推送方法、设备和介质

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108733703A (zh) * 2017-04-20 2018-11-02 北京京东尚科信息技术有限公司 问答系统的答案预测方法及装置、电子设备、存储介质
CN108766545A (zh) * 2018-05-31 2018-11-06 平安医疗科技有限公司 在线问诊科室分配方法、装置、计算机设备和存储介质
US20190272636A1 (en) * 2017-06-19 2019-09-05 Viz.ai, Inc. Method and system for computer-aided triage
CN111523976A (zh) * 2020-04-23 2020-08-11 京东数字科技控股有限公司 商品推荐方法、装置、电子设备及存储介质
CN111782774A (zh) * 2019-04-03 2020-10-16 北京嘀嘀无限科技发展有限公司 一种问题推荐的方法及装置
CN112287232A (zh) * 2020-11-09 2021-01-29 北京京东拓先科技有限公司 生成推荐信息的方法、装置

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108733703A (zh) * 2017-04-20 2018-11-02 北京京东尚科信息技术有限公司 问答系统的答案预测方法及装置、电子设备、存储介质
US20190272636A1 (en) * 2017-06-19 2019-09-05 Viz.ai, Inc. Method and system for computer-aided triage
CN108766545A (zh) * 2018-05-31 2018-11-06 平安医疗科技有限公司 在线问诊科室分配方法、装置、计算机设备和存储介质
CN111782774A (zh) * 2019-04-03 2020-10-16 北京嘀嘀无限科技发展有限公司 一种问题推荐的方法及装置
CN111523976A (zh) * 2020-04-23 2020-08-11 京东数字科技控股有限公司 商品推荐方法、装置、电子设备及存储介质
CN112287232A (zh) * 2020-11-09 2021-01-29 北京京东拓先科技有限公司 生成推荐信息的方法、装置

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115274086A (zh) * 2022-09-27 2022-11-01 无码科技（杭州）有限公司 一种智能导诊方法及系统
CN115687674A (zh) * 2022-12-20 2023-02-03 昆明勤砖晟信息科技有限公司 服务于智慧云服务平台的大数据需求分析方法及系统
CN116450925A (zh) * 2022-12-27 2023-07-18 深圳市网新新思软件有限公司 基于人工智能的用户关系分析方法及系统
CN116450925B (zh) * 2022-12-27 2023-12-15 深圳市网新新思软件有限公司 基于人工智能的用户关系分析方法及系统
CN116629810A (zh) * 2023-07-24 2023-08-22 中建五局第三建设有限公司 基于建筑办公系统的操作推荐方法、装置、设备和介质
CN116629810B (zh) * 2023-07-24 2023-10-20 中建五局第三建设有限公司 基于建筑办公系统的操作推荐方法、装置、设备和介质

Also Published As

Publication number Publication date
CN112287232B (zh) 2024-04-16
CN112287232A (zh) 2021-01-29

Similar Documents

Publication Publication Date Title
WO2022095893A1 (zh) 生成推荐信息的方法、装置
US11636847B2 (en) Ontology-augmented interface
CN112507715B (zh) 确定实体之间关联关系的方法、装置、设备和存储介质
US11645314B2 (en) Interactive information retrieval using knowledge graphs
US11232365B2 (en) Digital assistant platform
EP4006909B1 (en) Method, apparatus and device for quality control and storage medium
US20140019128A1 (en) Voice Based System and Method for Data Input
US11847411B2 (en) Obtaining supported decision trees from text for medical health applications
US10585902B2 (en) Cognitive computer assisted attribute acquisition through iterative disclosure
US20210056129A1 (en) Mapping of Topics within a Domain Based on Terms Associated with the Topics
CN112530576A (zh) 一种线上医患匹配方法、装置、电子设备及存储介质
CN114429133A (zh) 通过神经机器阅读理解依赖于话语分析来回答复杂问题
JP2022544047A (ja) 深層ｃｃａ及び能動ペアワイズクエリを用いるマルチモーダル検索及びクラスタリングのための方法
WO2022227203A1 (zh) 基于对话表征的分诊方法、装置、设备及存储介质
CN114360711A (zh) 通过句法-语义对准和话语分析进行的基于多案例的推理
CN111967599B (zh) 用于训练模型的方法、装置、电子设备及可读存储介质
US20240143605A1 (en) System And Method For Improved State Identification And Prediction In Computerized Queries
US20230082014A1 (en) Artificial Intelligence Based Technologies for Improving Patient Intake
US20240095445A1 (en) Systems and methods for language modeling with textual clincal data
JP2024512597A (ja) マルチモーダル談話ツリーを生成するための技術
FADHLALLAH A Deep Learning-based approach for Chatbot: medical assistance a case study
AU2022204665B2 (en) Automated search and presentation computing system
US20230043849A1 (en) Answer generation using machine reading comprehension and supported decision trees
US20240005231A1 (en) Methods and systems for holistic medical student and medical residency matching
KR20230174865A (ko) 채팅 상담 서비스를 제공하는 방법 및 장치

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21888590

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21888590

Country of ref document: EP

Kind code of ref document: A1