Information processing method, device, electronic device and readable storage medium

Abstract

The embodiments of the present application provide an information processing method, device, electronic device, and readable storage medium, which are intended to improve the efficiency of task execution. The information processing method is applied to task-oriented intelligent dialogues, and includes the following steps: determining one or more slot names required to complete the task based on the task in the dialogue message; selecting at least one candidate slot value for each of the one or more slot names based on at least one data source, wherein the slot name and each candidate slot value in the at least one candidate slot value are used as a predicted slot value pair; determining the confidence level of each of the at least one predicted slot value pairs corresponding to each slot name based on at least one round of dialogue messages; and generating a dialogue message based on the at least one predicted slot value pair corresponding to each slot name and its confidence level.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of intelligent interaction technology, and in particular to an information processing method, device, electronic device, and readable storage medium. Background Art

[0002] With the development of intelligent interaction technology and the popularization of intelligent electronic devices, more and more users are engaging in human-computer interaction through electronic devices and intelligent interaction systems. Electronic devices can serve as terminals, through which users interact online with intelligent interaction systems located on the server side. Alternatively, electronic devices can be equipped with intelligent interaction systems, allowing users to interact with them offline directly.

[0003] In the field of intelligent interaction, task-oriented intelligent dialogue technology is a popular technology. This technology aims to provide services to users by completing user-specified tasks through communication. For example, in the case of purchasing train tickets, current intelligent interaction systems will ask the user multiple rounds of questions during the interaction until they have obtained all the information provided by the user to recommend train tickets. For example, the interaction process between an intelligent interaction system and a user is as follows: User: Book me a train ticket. System: Where is your departure point? User: Shanghai. System: Where is your destination? User: Beijing. System: When do you leave? User: Tomorrow morning. System: What class of seat do you prefer? User: Second class. System: The available trains are: Train 1: 10:41 AM, train number G122, fare 553 yuan; Train 2: 11:05 AM, train number G126, fare 553 yuan; and so on.

[0004] It can be seen that the current intelligent interactive system lacks flexibility when processing information, has low efficiency in executing tasks, and affects the user experience. Summary of the Invention

[0005] The embodiments of the present application provide an information processing method, device, electronic device, and readable storage medium, aiming to improve task execution efficiency.

[0006] In a first aspect, an embodiment of the present application provides an information processing method, which is applied to task-oriented intelligent dialogue. The information processing method includes the following steps:

[0007] Based on the task in the conversation message, determine one or more slots required to complete the task;

[0008] selecting, based on at least one data source, at least one candidate slot value for each of the one or more slot names, wherein the slot name and each of the at least one candidate slot value are regarded as a predicted slot value pair;

[0009] Determining, based on at least one round of dialogue messages, a confidence level of at least one predicted slot-value pair corresponding to each slot name; and

[0010] Generate a dialog message based on at least one predicted slot-value pair corresponding to each slot name and its confidence.

[0011] Optionally, the method also includes a step of collecting information from the at least one data source, wherein the at least one data source includes: historical conversation information of the user in the current conversation, user portraits, user information collected by other conversation systems, historical information and real-time information of the user obtained by a third-party service, at least one of them.

[0012] Optionally, the step of selecting at least one candidate slot value for each slot name based on at least one data source includes:

[0013] Mapping information in at least one data source to the slot name to obtain a predicted slot-value pair; wherein the information in at least one data source is mapped to the corresponding slot name through a classification model.

[0014] Optionally, the classification model is obtained by training a convolutional neural network.

[0015] Optionally, the step of determining the confidence level of at least one predicted slot-value pair corresponding to each slot name based on at least one round of dialogue messages includes:

[0016] Determining a data source to which the slot value in the predicted slot value pair belongs;

[0017] The confidence level of the predicted slot-value pair is determined according to the at least one round of dialogue messages and the data source to which the slot value belongs.

[0018] Optionally, the step of determining the confidence level of at least one predicted slot-value pair corresponding to each slot name based on at least one round of dialogue messages includes:

[0019] Inputting the current round of dialogue message into a first feature extraction model to extract message features of the current round of dialogue message;

[0020] Inputting the message feature data and historical accumulated conversation features into a second feature extraction model to extract current accumulated conversation features for messages in a current round of conversation, wherein the historical accumulated conversation features are accumulated conversation features extracted from messages in a previous round of conversation by the second feature extraction model;

[0021] The confidence level of the predicted slot-value pair is determined based on the current accumulated conversation features and the slot-value pair features corresponding to the predicted slot-value pair.

[0022] Optionally, the step of inputting the current round of dialogue message into a first feature extraction model to extract message features of the current round of dialogue message includes:

[0023] Performing local feature extraction on the current round of dialogue messages to extract local features of the current round of dialogue messages;

[0024] Extracting global features of the current round of dialogue messages based on the current round of dialogue messages and historical global features, where the historical global features are global features extracted from the previous round of dialogue messages;

[0025] The message features of the current round of dialogue message are determined according to the local features and global features of the current round of dialogue message and their respective weights.

[0026] Optionally, the following steps are also included:

[0027] Determining a data source to which the slot value in the predicted slot value pair belongs;

[0028] The slot value pair feature corresponding to the predicted slot value pair is determined according to the data source to which the slot value belongs, the slot name and the slot value in the predicted slot value pair.

[0029] Optionally, the step of generating a dialog message according to at least one predicted slot value corresponding to each slot name and its confidence level includes:

[0030] comparing the confidence level of the at least one predicted slot-value pair with a preset threshold;

[0031] At least one predicted slot-value pair is selected from the at least one predicted slot-value pair greater than the preset threshold, and a dialog message of the slot name and slot value for the predicted slot-value pair is generated and output.

[0032] Optionally, the step of generating a dialog message according to at least one predicted slot value corresponding to each slot name and its confidence level includes:

[0033] Comparing the confidence of the predicted slot-value pair with a preset threshold;

[0034] In a case where the confidence level is greater than the preset threshold, determining the predicted slot value pair as a known slot value pair;

[0035] Determine a target slot name from the multiple slot names, wherein the target slot name is a slot name for which a corresponding slot value has not been determined;

[0036] Generate and output a dialog message for the target slot name.

[0037] A second aspect of an embodiment of the present application provides an information processing device for use in task-oriented intelligent dialogue, the information processing device comprising the following modules:

[0038] A slot name determination module is used to determine one or more slot names required to complete the task based on the task in the dialogue message;

[0039] a predicted slot value pair determination module, configured to select, based on at least one data source, at least one candidate slot value for each of the one or more slot names, wherein the slot name and each of the at least one candidate slot value are regarded as a predicted slot value pair;

[0040] a confidence determination module, configured to determine the confidence of at least one predicted slot-value pair corresponding to each slot name based on at least one round of dialogue messages; and

[0041] The dialog message generation module is used to generate a dialog message according to at least one predicted slot-value pair corresponding to each slot name and its confidence.

[0042] Optionally, the information processing device also includes an information collection module for collecting information from the at least one data source, wherein the at least one data source includes: historical conversation information of the user in the current conversation, user portraits, user information collected by other conversation systems, historical information and real-time information of the user obtained by a third-party service, at least one of them.

[0043] Optionally, the predicted slot value pair determination module includes:

[0044] A mapping submodule is used to map the information in at least one data source to the slot name to obtain a predicted slot-value pair; wherein the information in at least one data source is mapped to the corresponding slot name through a classification model.

[0045] Optionally, the classification model is obtained by training a convolutional neural network.

[0046] Optionally, the confidence determination module includes:

[0047] A first data source determination submodule, configured to determine a data source to which the slot value in the predicted slot value pair belongs;

[0048] The first confidence determination submodule is configured to determine the confidence of the predicted slot-value pair based on the at least one round of dialogue messages and the data source to which the slot value belongs.

[0049] Optionally, the confidence determination module includes:

[0050] a message feature extraction submodule, configured to input the current round of dialogue message into a first feature extraction model to extract message features of the current round of dialogue message;

[0051] a current cumulative conversation feature extraction submodule, configured to input the message feature data and historical cumulative conversation features into a second feature extraction model to extract current cumulative conversation features for the current round of conversation messages, where the historical cumulative conversation features are cumulative conversation features extracted from the previous round of conversation messages by the second feature extraction model;

[0052] The second confidence determination submodule is used to determine the confidence of the predicted slot-value pair based on the current accumulated conversation features and the slot-value pair features corresponding to the predicted slot-value pair.

[0053] Optionally, the message feature extraction submodule includes:

[0054] A local feature extraction subunit is used to extract local features of the current round of dialogue messages to extract local features of the current round of dialogue messages;

[0055] A global feature extraction subunit, configured to extract global features of the current round of dialogue messages based on the current round of dialogue messages and historical global features, wherein the historical global features are global features extracted from the previous round of dialogue messages;

[0056] The message feature determination subunit is configured to determine the message features of the current round of dialogue message according to the local features and global features of the current round of dialogue message and their respective weights.

[0057] Optionally, the confidence determination module further includes:

[0058] A second data source determination submodule, configured to determine a data source to which the slot value in the predicted slot value pair belongs;

[0059] The slot value pair feature determination submodule is used to determine the slot value pair feature corresponding to the predicted slot value pair based on the data source to which the slot value belongs, the slot name and the slot value in the predicted slot value pair.

[0060] Optionally, the dialogue message generation module includes:

[0061] a first confidence comparison module, configured to compare the confidence of the at least one predicted slot value pair with a preset threshold;

[0062] The first dialog message generating submodule is configured to select at least one predicted slot value pair from the at least one predicted slot value pair greater than the preset threshold, generate and output a dialog message of the slot name and slot value for the predicted slot value pair.

[0063] Optionally, the dialogue message generation module includes:

[0064] a second confidence comparison module, configured to compare the confidence of the predicted slot value pair with a preset threshold;

[0065] a known slot value pair determination module, configured to determine the predicted slot value pair as a known slot value pair when the confidence level is greater than the preset threshold;

[0066] a target slot name determining module, configured to determine a target slot name from the plurality of slot names, wherein the target slot name is a slot name for which a corresponding slot value has not been determined;

[0067] The second dialog message generating submodule is configured to generate and output a dialog message targeting the target slot name.

[0068] A third aspect of an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps in the method described in the first aspect of the present application are implemented.

[0069] A fourth aspect of an embodiment of the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executed, implements the steps of the method described in the first aspect of the present application.

[0070] Using the information processing method provided in the embodiment of the present application, the intelligent interactive system selects possible candidate slot values for each slot name from the data source for the slot name required to complete the user task, and forms a predicted slot value pair. Then, based on the dialogue message between the user and the system, the confidence level of each predicted slot value pair is determined. Finally, a dialogue message is generated based on each predicted slot value pair and its confidence level. Using this method, during the interaction with the user, the slot value corresponding to each slot name can be quickly determined while reducing the number of dialogue turns with the user. This method has high flexibility and is conducive to improving the efficiency of task execution and enhancing the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments of the present application. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0072] Figure 1 This is a schematic diagram of the structure of the first preset model and the classification model proposed in one embodiment of the present application;

[0073] Figure 2 Schematic diagram of the structure of the second preset model and the confidence prediction model proposed in one embodiment of the present application;

[0074] Figure 3 is a flow chart of an information processing method proposed in one embodiment of the present application;

[0075] Figure 4 It is a schematic diagram of an information processing device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0076] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are part of the embodiments of this application, not all of them. Based on the embodiments of this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0077] Task-oriented intelligent dialogue technology is a popular technology in the field of intelligent interaction. This technology aims to provide services to users by completing user-specified tasks through communication. For example, during the interaction process, current intelligent interaction systems first determine the slots required for the task, such as departure time, departure location, destination, and seat class. The intelligent interaction system then asks the user questions for each slot in turn. Only after the user has answered all the corresponding slot values can the system recommend the corresponding train ticket. This suggests that current intelligent interaction systems lack flexibility in information processing, resulting in low task execution efficiency and a negative impact on the user experience.

[0078] To improve task execution efficiency, the inventors of this application propose: for each slot name required to complete a user task, possible candidate slot values are selected from a data source to form predicted slot-value pairs. The confidence level of each predicted slot-value pair is then determined based on conversation messages with the user. Finally, a conversation message is generated based on each predicted slot-value pair and its confidence level.

[0079] Furthermore, to enable the intelligent interactive system or a device running the intelligent interactive system to more intelligently implement the aforementioned method proposed by the inventors, the inventors first constructed a first preset model, collected a first training sample set corresponding to the first preset model, and trained the first preset model based on the first training sample set to obtain a classification model. The inventors also constructed a second preset model, collected a second training sample set corresponding to the second preset model, and trained the second preset model based on the second training sample set to obtain a confidence prediction model.

[0080] The trained classification model and confidence prediction model can be used as part of an intelligent interactive system to execute all or part of the steps in the above method.

[0081] refer to Figure 1 , Figure 1 This is a schematic diagram of the structure of the first preset model and classification model proposed in one embodiment of the present application. Figure 1As shown, the first preset model mainly includes an input module 01, a convolution module 02, a pooling module 03 and an output module 04.

[0082] Among them, the input module 01 is mainly used to receive data source information and vectorize the data source information to obtain a vectorized representation of the data source information, Word Embedding. The convolution module 02 is mainly used to extract features from the vector of the data source information. The convolution module 02 can specifically use CNN (Convolutional Neural Networks). The convolution module 02 may include one or more convolution layers. The pooling module 03 is used to perform a pooling operation on the output of the convolution module. The pooling module 03 may include one or more pooling layers. The output module 04 can be implemented in a variety of ways. For example, the output module 04 may include at least one fully connected layer for receiving the output of the pooling module 03 and outputting the possibility, i.e., probability, of the data source information being classified into each slot name for each data source information.

[0083] For example, consider the user task of purchasing a train ticket. The slot names required for this task include departure time, departure location, destination, and seat class. Before training, a first training sample set can be collected. This first training sample set includes multiple first training samples, each of which contains a piece of information from a data source, labeled with a corresponding slot name. The data source can include historical user conversations, user profiles, user information collected by other conversation systems, and historical and real-time user information obtained from third-party services.

[0084] For example, a first training sample may be: the positioning software provides the current location "Beijing" and marks "departure place".

[0085] Or for example, a first training sample may be, for example: based on the user's historical conversation information, it is known that the user goes to Jinan on the 15th of every month, and the current time provided by the time software is the 14th, with "destination" and "departure time" marked.

[0086] Then, the first preset model is trained based on the collected first training sample set. During the training, a loss value can be determined based on the prediction result of the first preset model for each first training sample and the label of the first training sample, and the first preset model is updated based on the loss value. Finally, the first preset model that has undergone multiple rounds of training is determined as a classification model. The structure of the classification model is the same as that of the first preset model, such as Figure 1 As shown, the model parameters in this classification model are different from the initial parameters of the first classification model.

[0087] refer to Figure 2 , Figure 2Schematic diagram of the structure of the second preset model and the confidence prediction model proposed in one embodiment of the present application. Figure 2 As shown, the second preset model mainly includes: a first feature extraction model, a second feature extraction model, a slot-value pair feature extraction module, and a summary module.

[0088] The first feature extraction model is used to extract the message features of the current round of dialogue messages. Specifically, the first feature extraction model may include a convolutional network, a memory network, and a weighting module. The convolutional network may be a CNN (convolutional neural network), and the memory network may be an LSTM (long short-term memory network). The convolutional network is used to extract the local features r of the current round of dialogue messages. CNN The memory network extracts the global feature r of the current round of dialogue message based on the current round of dialogue message and historical global features. LSTM , where the historical global features are global features extracted from the previous round of dialogue messages. The weighting module is used to perform weighted calculation based on the local features and global features and their respective weights to determine the message feature r of the current round of dialogue messages.

[0089] The second feature extraction module can also use a network with memory function such as LSTM. The second feature extraction module is based on the message feature r of the current round of dialogue message and the historical accumulated dialogue feature h t-1 , extract the current cumulative dialogue feature h of the current round of dialogue messages t , where the historical cumulative dialogue feature h t-1 It is the cumulative conversation features extracted from the previous round of conversation messages.

[0090] The slot-value pair feature extraction module is used to convert the slot-value pair into a vector form and use the converted vector as the feature of the slot-value pair. Specifically, the slot-value pair feature extraction module may include a first conversion module, a second conversion module and a summation module. The first conversion module is used to convert the slot name and slot value source in the slot-value pair into a vector c slot The second conversion module is used to convert the slot value in the slot value pair into a vector c value The summation module is used to Figure 2 The formula shown is for c slot and c value Perform a sum operation to obtain the feature c of the slot-value pair.

[0091] The summary module is used to Figure 2 The formula shown is for the current cumulative conversation feature h t The feature c of the slot-value pair is summarized and calculated to obtain the slot-value pair confidence D.

[0092] Before training, a second training sample set can be collected in advance. This second training sample set includes multiple second training samples. The multiple second training samples can be divided into multiple groups. For each of the multiple groups, one or more second training samples in the group are derived from the same conversation. The conversation includes one or more rounds of conversation messages, and each round of conversation messages corresponds to a second training sample. Each second training sample includes: a conversation message, a slot-value pair, slot-value source information, and a pre-labeled confidence score.

[0093] Then, the second preset model is trained based on the collected second training sample set. During the training, the loss value can be determined based on the prediction confidence of each second training sample and the confidence of the second training sample label of the second preset model, and the second preset model is updated according to the loss value. Finally, the second preset model that has undergone multiple rounds of training is determined as the confidence prediction model. The structure of the confidence prediction model is the same as that of the second preset model, such as Figure 2 As shown, the model parameters in the confidence prediction model are different from the initial parameters of the second classification model.

[0094] refer to Figure 3 , Figure 3 This is a flow chart of an information processing method proposed in an embodiment of the present application, which is applied in task-oriented intelligent dialogue. Figure 3 As shown, the method includes the following steps:

[0095] Step S11: According to the task in the dialogue message, determine one or more slot names required to complete the task.

[0096] In this embodiment, the task refers to an action that the intelligent interactive system can assist the user to complete, such as assisting the user in purchasing train tickets, movie tickets, ordering food, booking a hotel, etc.

[0097] For example, step S11 may specifically include the following sub-steps:

[0098] Step S11 - 1 : Acquire a dialogue message, determine task information from the dialogue information, and determine the task corresponding to the dialogue message.

[0099] Step S11-2: Based on the determined task, all slot names required to complete the task are obtained, and based on the dialogue message, slot names with existing slot values are determined from the dialogue message. All slot names required to complete the task may be pre-set.

[0100] Step S11-3: Based on all slot names required to complete the task and the slot names with existing slot values, all slot names still required to complete the task are determined. In this embodiment, for an intelligent interactive system, the task of a conversation message can be fixed. For example, if the intelligent interactive system is specifically designed to assist users in purchasing train tickets, the task of the conversation message can be fixed as "Purchase train tickets." In this application scenario, the slot names required to complete the task can also be fixed, such as departure time, departure location, destination, and seat class.

[0101] In this embodiment, for an intelligent interactive system, the tasks of conversational messages can also be fluid. For example, an intelligent interactive system can assist users in completing various tasks, such as purchasing train tickets, movie tickets, recommending takeout, and booking hotels. The slot names corresponding to each task are typically different. For example, the slot names for the "Buy Train Tickets" task include departure time, departure location, destination, and seat class, while the slot names for the "Buy Movie Tickets" task include movie time, location, and movie title. In this application scenario, the intelligent interactive system needs to determine the user's target task type based on the conversational message input by the user.

[0102] For example, the intelligent interaction system can determine the user's target task type based on rules. For example, if the user's input dialogue message includes the entity "movie" or a movie name, the user's target task type can be determined to be "buying movie tickets." For example, if the user's input dialogue message includes the term "takeout" or a food category, the user's target task type can be determined to be "recommending takeout."

[0103] Alternatively, the intelligent interactive system can utilize a pre-trained task discrimination model to determine the user's target task type. This task discrimination model is trained and updated using user conversation messages labeled with task types as training samples. During use, the user's conversation messages are input into the task discrimination model to obtain the predicted task type output by the task discrimination model, which is then determined as the user's target task type.

[0104] Taking the task of "buying train tickets" in the conversation message as an example, the slot names required to complete the task include: departure time, departure place, destination, and seat class. In this embodiment, at a certain stage of the conversation, the intelligent interaction system may not have determined the slot value corresponding to any of the above slot names. Therefore, at this stage, the intelligent interaction system may determine all four of the above slot names as the slot names required to complete the task. Alternatively, at a certain stage of the conversation, the intelligent interaction system may have determined the slot values of some of the above four slot names, such as the slot values of the departure place and destination. Therefore, at this stage, the intelligent interaction system may determine the remaining slot names as the slot names required to complete the task, such as determining the departure time and seat class as the slot names required to complete the task.

[0105] Continuing with the above example, let's assume the user enters the message "Please book a train ticket for me." At this stage, the intelligent interaction system might determine all four slots as required to complete the task. Alternatively, let's assume the user enters the message "Please book a train ticket for tomorrow." At this stage, the intelligent interaction system might determine the departure location, destination, and seat class as required slots.

[0106] Continuing with the above example, assuming that after the first two rounds of dialogue, the intelligent interactive system has determined the slot values for the departure time, departure place, and destination, then at this stage, the intelligent interactive system may determine the seat grade as the slot name required to complete the task.

[0107] Step S12: Based on at least one data source, select at least one candidate slot value for each of the one or more slot names, wherein the slot name and each of the at least one candidate slot value are regarded as a predicted slot value pair.

[0108] In this embodiment, before step S12, information may be collected in advance from at least one data source. The at least one data source includes at least one of historical conversation information of the user in the current conversation, user profiles, user information collected by other conversation systems, and historical and real-time information of the user obtained by a third-party service.

[0109] The user's historical conversation information in the current conversation refers to the conversation messages between the user and the intelligent interaction system during previous tasks, including the user's input and the system's responses. For example, if the intelligent interaction system is specifically designed to assist users in purchasing train tickets, the historical tasks refer to the tasks that the user completed prior to the current task using the intelligent interaction system to purchase train tickets.

[0110] User portrait refers to user attribute information extracted from massive user data. This user attribute information may include one or more of the following information: basic attributes such as gender, occupation, age group, income level, marital status, and education level; behavioral attributes such as APP usage frequency and order probability; and interest attributes such as takeout preferences, movie preferences, and product preferences.

[0111] Other dialogue systems can be other intelligent interaction systems. For example, an intelligent interaction system that assists users in completing various tasks can obtain user information collected by an intelligent interaction system specifically designed to assist users in purchasing movie tickets, such as historical conversation information and user profiles, during its interaction with the user.

[0112] Third-party services can be other services running on the same device as the intelligent interactive system, or can be other services running on the same device as the client corresponding to the intelligent interactive system. For example, the client corresponding to the intelligent interactive system runs on the user's mobile phone, and the user's mobile phone also runs positioning services, weather forecast services, food delivery services, ticket ordering and other services, which can serve as the above-mentioned third-party services. For another example, "the user's current location is Beijing" is the user's real-time information provided by the third-party service (positioning service), "the user has checked the weather in Jinan tomorrow" is the user's historical information provided by the third-party service (weather forecast service), "the user ordered food delivery in the Wangjing area of Beijing two days ago" is the user's historical information provided by the third-party service (food delivery service), and "the user booked a train ticket from Beijing to Jinan on August 15th" is the user's historical information provided by the third-party service (ticket ordering service).

[0113] In this embodiment, in order to select candidate slot values, a feasible implementation method is: mapping the information in at least one data source to the slot name to obtain a predicted slot value pair; wherein, the information in at least one data source is mapped to the corresponding slot name through a classification model.

[0114] For example, the information in the data source can be entered into Figure 1The classification model shown in Figure 1 converts information into a vector form, such as a word embedding, through an input module. It then passes through a convolutional module, a pooling module, and an output module, outputting the probability of the information being classified into each slot. Taking the task of purchasing train tickets as an example, assuming the input information is the location information "the user's current location is Beijing" provided by positioning service software (a third-party service), the classification model output can be a probability vector, for example, (0.08, 0.74, 0.16, 0.02), where the four probabilities correspond to the four slots: departure time, departure place, destination, and seat class. In this case, "Departure place: Beijing" can be determined as one predicted slot value pair, or "Departure place: Beijing" and "Destination: Beijing" can be determined as two predicted slot value pairs.

[0115] In the above implementation, the classification model is obtained by training a convolutional neural network. Figure 1 As shown, Figure 1 The classification model shown includes: Convolutional Neural Network module.

[0116] In addition, in this embodiment, in order to select a candidate slot value, another feasible implementation method is: determining a candidate slot value for a slot name based on a rule.

[0117] For example, when purchasing a train ticket, if the data source is the current location provided by the location service software, the location represented by the current location is determined as a candidate slot value for the departure location according to the rules. Alternatively, if the data source is seat class information included in historical conversations, the class category represented by the seat class information is determined as a candidate slot value for the seat class according to the rules.

[0118] Step S13: Determine the confidence level of at least one predicted slot-value pair corresponding to each slot name based on at least one round of dialogue messages.

[0119] In this embodiment, the confidence level of a predicted slot-value pair refers to the degree of trustworthiness of the predicted slot-value pair. A higher confidence level indicates a greater likelihood that the slot name and slot value in the predicted slot-value pair correspond to each other; a lower confidence level indicates a lower likelihood that the slot name and slot value in the predicted slot-value pair correspond to each other.

[0120] In this embodiment, in order to determine the confidence level of the predicted slot value pair, step S13 may specifically include the following steps:

[0121] Step S13-1: Determine the data source to which the slot value in the predicted slot value pair belongs;

[0122] Step S13 - 2 : Determine the confidence level of the predicted slot-value pair based on the at least one round of dialogue messages and the data source to which the slot value belongs.

[0123] The dialogue message includes a dialogue message input by the user and / or a response message output by the intelligent interactive system.

[0124] Taking the train ticket purchase task as an example, assume that step S12 determines two predicted slot value pairs for the departure location: "Departure: Beijing" and "Departure: Jinan." The slot value "Beijing" is derived from location information provided by the positioning service software (a third-party service), while the slot value "Jinan" is derived from the user-entered message "Departing from Jinan" in historical conversations. Confidence can be determined not only based on existing conversations between the user and the intelligent interactive system, but also based on the source of the slot value.

[0125] When the source of the slot value is used as the basis, one feasible implementation is to determine, for each slot name, multiple data sources that may correspond to that slot name, and then sort these data sources by their credibility, with higher-ranked data sources having higher influence coefficients. After determining the confidence level of a predicted slot-value pair based on at least one round of conversation messages, the confidence level is multiplied by the influence coefficient corresponding to the data source to which the slot value in the predicted slot-value pair belongs, to obtain the final confidence level.

[0126] Continuing with the above example, let's assume that the multiple data sources corresponding to the slot name "Departure Place" are ranked by credibility as follows: real-time location provided by the location service software, location information from the weather forecast software, the user's active cities in the user profile, and the user's historical conversation information. Based on this ranking, it can be determined that for the two predicted slot value pairs "Departure Place: Beijing" and "Departure Place: Jinan," the data source corresponding to the slot value "Beijing" has a higher influence coefficient, while the data source corresponding to the slot value "Jinan" has a lower influence coefficient. Based on the influence coefficients corresponding to each data source and at least one round of conversation messages, the final confidence level for each of these two predicted slot value pairs can be determined.

[0127] In this embodiment, the data source to which the slot value belongs is taken into consideration when executing the above steps S13-1 and S13-2. Taking the data source as a factor affecting the confidence level can effectively improve the accuracy of the confidence level and further enhance the user experience.

[0128] In addition, in this embodiment, in order to determine the confidence level of the predicted slot value pair, step S13 may further specifically include the following steps:

[0129] Step S13-A: Input the current round of dialogue message into a first feature extraction model to extract message features of the current round of dialogue message.

[0130] Step S13-B: Input the message feature data and historical cumulative conversation features into a second feature extraction model to extract the current cumulative conversation features for the current round of conversation messages. The historical cumulative conversation features are the cumulative conversation features extracted from the previous round of conversation messages by the second feature extraction model.

[0131] Step S13-C: Determine the confidence level of the predicted slot-value pair based on the current accumulated conversation features and the slot-value pair features corresponding to the predicted slot-value pair.

[0132] The dialogue message includes the dialogue message input by the user and / or the response message output by the intelligent interactive system. Figure 2 The first feature extraction model in the confidence prediction model shown. Figure 2 The first feature extraction model in includes a convolutional network, a memory network and a weighted module. Thus, step S13-A may further specifically include:

[0133] Step S13-A-1: Perform local feature extraction on the current round of dialogue messages to extract local features of the current round of dialogue messages.

[0134] Step S13-A-2: extracting the global features of the current round of dialogue messages based on the current round of dialogue messages and the historical global features, where the historical global features are global features extracted from the previous round of dialogue messages.

[0135] Step S13-A-3: Determine the message features of the current round of dialogue message according to the local features and global features of the current round of dialogue message and their respective weights.

[0136] Specifically, first convert the current round of dialogue messages into vector form, such as word embedding. Then input the word embeddings into Figure 2 The convolutional network and memory network in the current round of dialogue are used to extract local features of the current round of dialogue messages to obtain the local features r of the current round of dialogue messages. CNN . Use the memory network to extract the global features of the current round of dialogue messages and obtain the global features r of the current round of dialogue messages. LSTM Finally, the local feature r CNN and global feature r LSTM Input weighting module, weighting module according to the formula: r = w1 × r LSTM +w2×r CNN The message feature r of the current round of dialogue message is determined. The parameters w1 and w2 in the above formula can be parameters that are gradually fixed during the training of the second preset model, or can be fixed parameters preset by technicians and not updated during the model training.

[0137] By extracting the message feature r in the manner of steps S13-A-1 to S13-A-3, not only the current local features are taken into account, but also the global features since the user and the intelligent interactive system started interacting, so that the extracted message feature r can more accurately represent the dialogue state.

[0138] For step S13-B, the message feature data r can be input into Figure 2 The second feature extraction model in the confidence prediction model shown in FIG. Specifically, due to Figure 2 The second feature extraction model in the example is the long short-term memory network LSTM (Long Short-Term Memory). When the second feature extraction model takes the message feature data r as input, it also uses the cumulative conversation feature h extracted from the previous round of conversation messages. t-1 The second feature extraction model outputs the current cumulative dialogue feature h for the current round of dialogue messages based on the above two inputs. t .

[0139] For step S13-C, the current accumulated conversation feature h t And the slot value pair feature c corresponding to the predicted slot value pair, input Figure 2 The summary module in the confidence prediction model shown in FIG. The summary module is sequentially calculated according to the formula d=h t ×c and D=sigmoid(d) determine the confidence D. The sigmoid() function is an S-shaped function, and the confidence D finally determined is a value between 0 and 1.

[0140] In addition, before step S13-C, the data source to which the slot value in the predicted slot value pair belongs can be determined in advance; and based on the data source to which the slot value belongs, the slot name and slot value in the predicted slot value pair, the slot value pair features corresponding to the predicted slot value pair can be determined.

[0141] For example, Figure 2 As shown, you can enter the data source and slot name to which the slot value belongs in the predicted slot value pair Figure 2 The first conversion module in , thereby obtaining the vectorized representation of the data source and slot name c slot For example, the data source and slot name can be converted into word vectors through the first conversion module. Figure 2 As shown, the slot value in the predicted slot value pair can be input into Figure 2 The second conversion module in , thereby obtaining the vectorized representation c of the slot value value For example, the slot value can be converted into a word vector through the second conversion module. Then according to the formula c=w c ×(c slot +c value )+wb Determine the slot value pair feature c corresponding to the predicted slot value pair. The parameter w in the above formula c and w b The parameters may be gradually fixed during the training of the second preset model; or they may be fixed parameters preset by technicians and not updated during the model training.

[0142] By executing steps S13-A through S13-C above, the confidence level of the predicted slot-value pair is determined. This factor, by considering the data source to which the slot value belongs and considering it as a factor influencing the confidence level, can effectively improve the accuracy of the confidence level and further enhance the user experience. Furthermore, by automatically determining the confidence level of the predicted slot-value pair using a pre-trained confidence prediction model, the intelligence of the intelligent interactive system can be enhanced, further improving the accuracy of the system's response.

[0143] Step S14: Generate a dialog message based on at least one predicted slot-value pair corresponding to each slot name and its confidence level.

[0144] In this embodiment, the dialogue messages generated by the intelligent interactive system can be text messages, voice messages, links recommended to users (such as merchant links, movie ticket ordering page links, etc.), and responses to system actions (such as automatic jump to the recommendation page, automatic phone calls, automatic text messages, etc.). This application does not limit the specific form of the dialogue messages generated by the intelligent interactive system.

[0145] Regarding step S14, a feasible specific implementation method includes the following steps:

[0146] Step S14-1: comparing the confidence level of the at least one predicted slot value pair with a preset threshold;

[0147] Step S14 - 2 : selecting at least one predicted slot-value pair from the at least one predicted slot-value pair greater than the preset threshold, generating and outputting a dialog message of the slot name and slot value for the predicted slot-value pair.

[0148] In step S14-2, all predicted slot value pairs with confidence levels greater than a preset threshold may be selected, and a dialog message may be generated and output for the selected predicted slot value pairs. Alternatively, the predicted slot value pair with the highest confidence level among all predicted slot value pairs with confidence levels greater than the preset threshold may be selected, and a dialog message may be generated and output for the selected predicted slot value pair.

[0149] For example, assume that for the current round of conversation messages, through steps S11 to S13, the corresponding confidence levels of the following two predicted slot value pairs are determined: "Departure Place: Beijing" and "Departure Place: Jinan." The confidence level for "Departure Place: Beijing" is 0.82, and the confidence level for "Departure Place: Jinan" is 0.22. Furthermore, assuming a preset threshold of 0.6, "Departure Place: Beijing" can be selected from the two predicted slot value pairs.

[0150] When generating a dialog message for a selected predicted slot-value pair, the slot name and slot value in the predicted slot-value pair can be output, allowing the user to confirm whether the predicted slot value for the slot name is accurate. For example, assuming the selected slot-value pair is "Departure Place: Beijing," the dialog message generated by the intelligent interactive system might be, "Excuse me, is your departure place Beijing?" Alternatively, assuming the selected slot-value pairs include "Departure Place: Beijing" and "Departure Place: Jinan," the dialog message generated by the intelligent interactive system might be, "Excuse me, is your departure place: 1. Beijing; 2. Jinan?"

[0151] In step S14-2, if the confidence levels of all predicted slot-value pairs are not greater than a preset threshold, a query may be directly sent to the user regarding the slot names of these predicted slot-value pairs, so that the user can input the slot value for the slot name.

[0152] Furthermore, if there are multiple prediction slot value pairs with confidence levels greater than a preset threshold, the confidence differences among these prediction slot value pairs with confidence levels greater than the preset threshold can be compared to determine whether to select one or more prediction slot value pairs from these prediction slot value pairs, and then generate a dialog message for the selected prediction slot value pairs. When the confidence difference is large, the prediction slot value pair with the highest confidence level can be selected from these prediction slot value pairs. When the confidence difference is small, two or more prediction slot value pairs with higher confidence levels can be selected from these prediction slot value pairs.

[0153] For example, assuming that for the current round of dialogue messages, through the aforementioned steps S11 to S13, the corresponding confidence levels of the following four predicted slot value pairs are determined: "Place of departure: Beijing", "Place of departure: Jinan", "Place of departure: Shanghai" and "Place of departure: Chengdu".

[0154] Among them, the confidence level corresponding to "Departure Place: Beijing" is 0.82, the confidence level corresponding to "Departure Place: Jinan" is 0.22, the confidence level corresponding to "Departure Place: Shanghai" is 0.79, and the confidence level corresponding to "Departure Place: Chengdu" is 0.66. Assuming that the preset threshold is 0.6, the confidence levels of the three prediction slot value pairs "Departure Place: Beijing", "Departure Place: Shanghai" and "Departure Place: Chengdu" among the above four prediction slot value pairs are greater than the preset threshold of 0.6. By comparing the confidence level differences between the three, assuming that the preset difference threshold is equal to 0.05, the confidence level difference between "Departure Place: Beijing" and "Departure Place: Shanghai" is 0.03, which is less than the difference threshold of 0.05. The confidence level difference between "Departure Place: Shanghai" and "Departure Place: Chengdu" is equal to 0.13, which is greater than the difference threshold of 0.05. In this way, the two prediction slot value pairs of "Departure place: Beijing" and "Departure place: Shanghai" are selected, and a dialogue message is generated for the two selected prediction slot value pairs, such as "Is your departure place Beijing or Shanghai?"

[0155] By using the above method to generate dialogue messages, when the confidence difference is small, dialogue messages are generated for multiple predicted slot value pairs with small differences, which can enhance the fault tolerance of the intelligent interactive system and further improve the user experience. Regarding step S14, another feasible specific implementation method includes the following steps:

[0156] Step S14-A: comparing the confidence of the predicted slot value pair with a preset threshold;

[0157] Step S14-B: when the confidence level is greater than the preset threshold, determining the predicted slot value pair as a known slot value pair;

[0158] Step S14-C: determining a target slot name from the multiple slot names, wherein the target slot name is a slot name for which a corresponding slot value has not been determined;

[0159] Step S14-D: Generate and output a dialog message for the target slot name.

[0160] Taking the purchase of train tickets as an example, the slot names required to complete this task include: departure time, departure place, destination, and seat class. Assume that after the user completes the first round of dialogue with the intelligent interactive system, through steps S11 to S14-B, it is determined that the confidence of the predicted slot value pairs such as "departure time: tomorrow", "departure place: Beijing", and "destination: Jinan" is greater than the preset threshold, and these three predicted slot value pairs are determined as known slot value pairs. Then in step S14-C, the intelligent interactive system determines "seat class" as the target slot name from the four required slot names, and generates a dialogue message for "seat class" in step S14-D. For example, an inquiry message is generated for "seat class": "Would you like to choose first class or second class?"

[0161] Furthermore, if, in step S14-B, the confidence levels of multiple predicted slot-value pairs for the same slot name are all greater than a preset threshold, the predicted key-value pair with the highest confidence level is determined as a known slot-value pair. For example, if the confidence levels of the two predicted slot-value pairs "Departure Place: Beijing" and "Departure Place: Jinan" are both greater than the preset threshold, and the confidence level of "Departure Place: Beijing" is greater than that of "Departure Place: Jinan", then "Departure Place: Beijing" can be determined as a known slot-value pair.

[0162] It should be understood that the above two specific implementation examples are only two examples among the multiple implementation methods of this application. As for the specific method used to generate the dialogue message in step S14, this application does not limit this.

[0163] By executing the information processing method described above, including steps S11 through S14, the intelligent interactive system selects possible candidate slot values for each slot name required to complete a user task from the data source and forms predicted slot-value pairs. The system then determines the confidence level of each predicted slot-value pair based on the conversational messages exchanged with the user. Finally, a conversational message is generated based on each predicted slot-value pair and its confidence level. This method allows for rapid determination of the slot value corresponding to each slot name while reducing the number of conversational turns with the user. This method offers high flexibility, improving task execution efficiency and enhancing the user experience.

[0164] Based on the same inventive concept, an embodiment of the present application provides an information processing device. Figure 4 , Figure 4 Schematic diagram of an information processing device provided by an embodiment of the present application. Figure 4 As shown, the device includes:

[0165] The slot name determination module 41 is used to determine one or more slot names required to complete the task according to the task in the dialogue message;

[0166] a predicted slot value pair determining module 42, configured to select at least one candidate slot value for each of the one or more slot names based on at least one data source, wherein the slot name and each candidate slot value of the at least one candidate slot value are regarded as a predicted slot value pair;

[0167] A confidence determination module 43 is configured to determine the confidence of at least one predicted slot-value pair corresponding to each slot name based on at least one round of dialogue messages; and

[0168] The dialogue message generation module 44 is configured to generate a dialogue message according to at least one predicted slot value and its confidence level corresponding to each slot name.

[0169] Optionally, the information processing device also includes an information collection module for collecting information from the at least one data source, wherein the at least one data source includes: historical conversation information of the user in the current conversation, user portraits, user information collected by other conversation systems, historical information and real-time information of the user obtained by a third-party service, at least one of them.

[0170] Optionally, the predicted slot value pair determination module includes:

[0171] A mapping submodule is used to map the information in at least one data source to the slot name to obtain a predicted slot-value pair; wherein the information in at least one data source is mapped to the corresponding slot name through a classification model.

[0172] Optionally, the classification model is obtained by training a convolutional neural network.

[0173] Optionally, the confidence determination module includes:

[0174] A first data source determination submodule, configured to determine a data source to which the slot value in the predicted slot value pair belongs;

[0175] The first confidence determination submodule is configured to determine the confidence of the predicted slot-value pair based on the at least one round of dialogue messages and the data source to which the slot value belongs.

[0176] Optionally, the confidence determination module includes:

[0177] a message feature extraction submodule, configured to input the current round of dialogue message into a first feature extraction model to extract message features of the current round of dialogue message;

[0178] a current cumulative conversation feature extraction submodule, configured to input the message feature data and historical cumulative conversation features into a second feature extraction model to extract current cumulative conversation features for the current round of conversation messages, where the historical cumulative conversation features are cumulative conversation features extracted from the previous round of conversation messages by the second feature extraction model;

[0179] The second confidence determination submodule is used to determine the confidence of the predicted slot-value pair based on the current accumulated conversation features and the slot-value pair features corresponding to the predicted slot-value pair.

[0180] Optionally, the message feature extraction submodule includes:

[0181] A local feature extraction subunit is used to extract local features of the current round of dialogue messages to extract local features of the current round of dialogue messages;

[0182] A global feature extraction subunit, configured to extract global features of the current round of dialogue messages based on the current round of dialogue messages and historical global features, wherein the historical global features are global features extracted from the previous round of dialogue messages;

[0183] The message feature determination subunit is configured to determine the message features of the current round of dialogue message according to the local features and global features of the current round of dialogue message and their respective weights.

[0184] Optionally, the confidence determination module further includes:

[0185] A second data source determination submodule, configured to determine a data source to which the slot value in the predicted slot value pair belongs;

[0186] The slot value pair feature determination submodule is used to determine the slot value pair feature corresponding to the predicted slot value pair based on the data source to which the slot value belongs, the slot name and the slot value in the predicted slot value pair.

[0187] Optionally, the dialogue message generation module includes:

[0188] a first confidence comparison module, configured to compare the confidence of the at least one predicted slot value pair with a preset threshold;

[0189] The first dialog message generating submodule is configured to select at least one predicted slot value pair from the at least one predicted slot value pair greater than the preset threshold, generate and output a dialog message of the slot name and slot value for the predicted slot value pair.

[0190] Optionally, the dialogue message generation module includes:

[0191] a second confidence comparison module, configured to compare the confidence of the predicted slot value pair with a preset threshold;

[0192] a known slot value pair determination module, configured to determine the predicted slot value pair as a known slot value pair when the confidence level is greater than the preset threshold;

[0193] a target slot name determining module, configured to determine a target slot name from the plurality of slot names, wherein the target slot name is a slot name for which a corresponding slot value has not been determined;

[0194] The second dialog message generating submodule is configured to generate and output a dialog message targeting the target slot name.

[0195] Based on the same inventive concept, another embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps in the method described in any of the above embodiments of the present application are implemented.

[0196] Based on the same inventive concept, another embodiment of the present application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executed, implements the steps of the method described in any of the above embodiments of the present application.

[0197] As for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0198] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0199] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, devices, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0200] The embodiments of the present application are described with reference to the flowcharts and / or block diagrams of the methods, terminal devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminal device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0201] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer readable memory produce a manufactured product including an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0202] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device so that a series of operating steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable terminal device to implement the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0203] Although preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they become aware of the basic inventive concepts. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the embodiments of the present invention.

[0204] Finally, it should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or terminal device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or terminal device that includes the element.

[0205] The above is a detailed introduction to the information processing method, device, electronic device and readable storage medium provided by the present application. Specific examples are used herein to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea. At the same time, for those skilled in the art, according to the ideas of the present application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. An information processing method, applied to task-oriented intelligent dialogue, characterized in that: The information processing method comprises the following steps: Based on the task in the conversation message, determine one or more slots required to complete the task; selecting, based on at least one data source, at least one candidate slot value for each of the one or more slot names, wherein the slot name and each of the at least one candidate slot value are regarded as a predicted slot value pair; Determining, based on at least one round of dialogue messages, a confidence level of at least one predicted slot-value pair corresponding to each slot name; and generating a dialog message based on at least one predicted slot-value pair corresponding to each slot name and its confidence score; The step of determining the confidence of at least one predicted slot-value pair corresponding to each slot name based on at least one round of dialogue messages includes: Inputting the current round of dialogue message into a first feature extraction model to extract message features of the current round of dialogue message; Inputting the message feature data and historical accumulated conversation features into a second feature extraction model to extract current accumulated conversation features for messages in a current round of conversation, wherein the historical accumulated conversation features are accumulated conversation features extracted from messages in a previous round of conversation by the second feature extraction model; Determining the confidence of the predicted slot-value pair based on the current accumulated conversation features and the slot-value pair features corresponding to the predicted slot-value pair; The step of inputting the current round of dialogue message into the first feature extraction model to extract the message features of the current round of dialogue message includes: Performing local feature extraction on the current round of dialogue messages to extract local features of the current round of dialogue messages; Extracting global features of the current round of dialogue messages based on the current round of dialogue messages and historical global features, where the historical global features are global features extracted from the previous round of dialogue messages; The message features of the current round of dialogue message are determined according to the local features and global features of the current round of dialogue message and their respective weights.

2. The information processing method according to claim 1, wherein: The method further includes the step of collecting information from the at least one data source, wherein the at least one data source includes at least one of historical conversation information of the user in the current conversation, user portraits, user information collected by other conversation systems, historical information and real-time information of the user obtained by a third-party service.

3. The information processing method according to claim 1, wherein: The step of selecting at least one candidate slot value for each slot name based on at least one data source includes: Mapping information in at least one data source to the slot name to obtain a predicted slot-value pair; wherein the information in at least one data source is mapped to the corresponding slot name through a classification model.

4. The information processing method according to claim 3, wherein: The classification model is obtained by training a convolutional neural network.

5. The information processing method according to claim 1, wherein: The step of determining the confidence of at least one predicted slot-value pair corresponding to each slot name based on at least one round of dialogue messages includes: Determining a data source to which the slot value in the predicted slot value pair belongs; The confidence level of the predicted slot-value pair is determined according to the at least one round of dialogue messages and the data source to which the slot value belongs.

6. The information processing method according to claim 1, wherein: The following steps are also included: Determining a data source to which the slot value in the predicted slot value pair belongs; The slot value pair feature corresponding to the predicted slot value pair is determined according to the data source to which the slot value belongs, the slot name and the slot value in the predicted slot value pair.

7. The information processing method according to any one of claims 1 to 6, characterized in that: The step of generating a dialog message according to at least one predicted slot-value pair corresponding to each slot name and its confidence level includes: comparing the confidence level of the at least one predicted slot-value pair with a preset threshold; At least one predicted slot-value pair is selected from the at least one predicted slot-value pair greater than the preset threshold, and a dialog message of the slot name and slot value for the predicted slot-value pair is generated and output.

8. The information processing method according to any one of claims 1 to 6, characterized in that: The step of generating a dialog message according to at least one predicted slot-value pair corresponding to each slot name and its confidence level includes: Comparing the confidence of the predicted slot-value pair with a preset threshold; In a case where the confidence level is greater than the preset threshold, determining the predicted slot value pair as a known slot value pair; Determine a target slot name from the multiple slot names, wherein the target slot name is a slot name for which a corresponding slot value has not been determined; Generate and output a dialog message for the target slot name.

9. An information processing device, applied to task-oriented intelligent dialogue, characterized in that: The information processing device includes the following modules: A slot name determination module is used to determine one or more slot names required to complete the task based on the task in the dialogue message; a predicted slot value pair determination module, configured to select at least one candidate slot value for each of the one or more slot names based on at least one data source, wherein the slot name and each candidate slot value of the at least one candidate slot value are regarded as a predicted slot value pair; a confidence determination module, configured to determine the confidence of at least one predicted slot-value pair corresponding to each slot name based on at least one round of dialogue messages; and a dialogue message generation module, configured to generate a dialogue message based on the at least one predicted slot-value pair corresponding to each slot name and its confidence; Determining the confidence of at least one predicted slot-value pair corresponding to each slot name according to at least one round of dialogue messages includes: Inputting the current round of dialogue message into a first feature extraction model to extract message features of the current round of dialogue message; Inputting the message feature data and historical accumulated conversation features into a second feature extraction model to extract current accumulated conversation features for messages in a current round of conversation, wherein the historical accumulated conversation features are accumulated conversation features extracted from messages in a previous round of conversation by the second feature extraction model; Determining the confidence of the predicted slot-value pair based on the current accumulated conversation features and the slot-value pair features corresponding to the predicted slot-value pair; Inputting the current round of dialogue message into the first feature extraction model to extract the message features of the current round of dialogue message includes: Performing local feature extraction on the current round of dialogue messages to extract local features of the current round of dialogue messages; Extracting global features of the current round of dialogue messages based on the current round of dialogue messages and historical global features, where the historical global features are global features extracted from the previous round of dialogue messages; The message features of the current round of dialogue message are determined according to the local features and global features of the current round of dialogue message and their respective weights.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

11. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When executed, the processor implements the steps of the method according to any one of claims 1 to 8.

Images