Voice data warehouse construction method and device, storage medium and electronic equipment

Abstract

The application provides a voice data warehouse construction method and device, a storage medium and an electronic device. The method comprises the following steps: receiving voice data, and extracting a business activity event in the voice data as a target business activity event; matching the target business activity event with a first correspondence relationship generated in advance to obtain a business information domain corresponding to the target business activity event as a target business information domain; matching the target business information domain with a second correspondence relationship generated in advance to obtain dimension information corresponding to the target business information domain as target dimension information; extracting an analysis object corresponding to the target dimension information in the voice data as to-be-stored information of a voice data warehouse, and constructing the voice data warehouse. In the application, the business abstract information is used as the analysis object, and the voice data warehouse construction efficiency and flexibility are improved.

Description

Technical Field

[0001] This application relates to the field of voice information processing technology, and in particular to a method, apparatus, storage medium and electronic device for constructing a voice data warehouse. Background Technology

[0002] The existing voice data warehouse is divided into a data source layer, a data detail layer, a common dimension layer, and a data mart layer. When voice data is input into the warehouse's input interface, it is processed in the following order: the voice data is saved to the data source layer; in the data detail layer, detailed data is extracted based on the voice data requirements; in the common dimension layer, the extracted detailed data is split into different data tables; and in the common dimension layer, summary statistics are performed based on the above data tables. The most basic premise of this voice data warehouse implementation is that it analyzes based on requirements. This approach has low reusability; each time new voice data is received, the entire process must be repeated. This results in poor flexibility and low resource utilization when dealing with large amounts of data. Summary of the Invention

[0003] The technical problem to be solved by this application is the poor flexibility and low resource utilization caused by the demand-driven approach in the construction of existing voice data warehouses. Therefore, this application provides a voice data warehouse construction method, device, storage medium and electronic device.

[0004] Firstly, the technical solution of this application provides a method for constructing a voice data warehouse, including:

[0005] Receive voice data and extract the business activity events in the voice data as target business activity events;

[0006] The target business activity event is matched with a pre-generated first correspondence to obtain the business information domain corresponding to the target business activity event as the target business information domain; wherein, the first correspondence is the correspondence between the business activity event and the business information domain; wherein, each business activity event is an indivisible behavioral event, and each business information domain is a preset function extracted from historical business activity events.

[0007] The target business information domain is matched with a pre-generated second correspondence to obtain the dimension information corresponding to the target business information domain as the target dimension information; wherein, the second correspondence is the correspondence between the business information domain and the dimension information, and each of the dimension information is an analysis object extracted from the historical business information domain;

[0008] Extract the analysis objects corresponding to the target dimension information from the speech data as the information to be stored in the speech data warehouse, and construct the speech data warehouse.

[0009] Some solutions describe a method for constructing a voice data warehouse, wherein receiving voice data and extracting business activity events from the voice data as target business activity events includes:

[0010] Convert the voice data into text information;

[0011] The text information is input into a preset semantic model to obtain the business activity event represented by the text information; the preset semantic model is trained by taking text information samples as input and the business activity events corresponding to the text information samples as output.

[0012] The business activity event represented by the text information is taken as the target business activity event.

[0013] In some solutions for constructing a voice data warehouse, the method involves matching the target business activity event with a pre-generated first correspondence to obtain the business information domain corresponding to the target business activity event as the target business information domain. The first correspondence is obtained in the following way:

[0014] Obtain application process runtime data for historical business activity events;

[0015] Extract the last running application process recorded in the application process running data as the target process, take the function to be implemented by the target process as the preset function, and take the preset function as the name of the business information domain;

[0016] The correspondence between the historical business activity events and the business information domain is taken as the first correspondence.

[0017] Some solutions describe methods for constructing voice data warehouses, where the last running application process recorded in the application process runtime data is extracted as the target process, and the function to be implemented by the target process is defined as the preset function. This includes:

[0018] When the target process is the same as the previously running multi-step application process, the preset function is multi-turn dialogue, which corresponds to the business information domain being a multi-turn dialogue domain.

[0019] In some solutions for constructing a voice data warehouse, the method involves matching the target business information domain with a pre-generated second correspondence to obtain the dimension information corresponding to the target business information domain, which is then used as the target dimension information.

[0020] The dimensional information in the second correspondence includes multi-level dimensional information; wherein, the dimensional information at the next level is obtained by dividing the dimensional information at the previous level according to a set classification method;

[0021] The target dimension information includes multi-level target dimension information obtained by matching the target business information domain with the second correspondence.

[0022] Some solutions describe methods for constructing a voice data warehouse, including the step of extracting the analysis object corresponding to the target dimension information from the voice data as the information to be stored in the voice data warehouse.

[0023] The analysis objects corresponding to each level of target dimension information in the voice data are divided into a group of information to be stored.

[0024] The different groups of information to be stored are stored in different storage areas of the voice data warehouse.

[0025] Secondly, the technical solution of this application provides a voice data warehouse construction device, characterized in that it includes:

[0026] The voice data receiving module is configured to receive voice data and extract business activity events from the voice data as target business activity events;

[0027] The first matching module is configured to match the target business activity event with a pre-generated first correspondence relationship to obtain a business information domain corresponding to the target business activity event as the target business information domain; wherein, the first correspondence relationship is the correspondence relationship between business activity events and business information domains; wherein, each business activity event is an indivisible behavioral event, and each business information domain is a preset function extracted from historical business activity events;

[0028] The second matching module is configured to match the target business information domain with a pre-generated second correspondence relationship to obtain the dimension information corresponding to the target business information domain as the target dimension information; wherein, the second correspondence relationship is the correspondence relationship between the business information domain and the dimension information, and each of the dimension information is an analysis object extracted from the historical business information domain;

[0029] The construction module is configured to extract the analysis objects corresponding to the target dimension information in the speech data as the information to be stored in the speech data warehouse, and to construct the speech data warehouse.

[0030] Some solutions describe a voice data warehouse construction apparatus in which the construction module is located in the data application layer of the operating system.

[0031] Thirdly, the present application provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the voice data warehouse construction method described in any of the first aspects.

[0032] Fourthly, the present application provides a computer-readable storage medium storing program information, wherein a computer reads the program information and executes the voice data warehouse construction method described in any one of the first aspects.

[0033] Fifthly, this application provides an electronic device, which includes at least one processor and at least one memory, wherein at least one memory stores program information, and at least one processor reads the program information and executes the voice data warehouse construction method according to any one of the first aspects.

[0034] The above technical solution has the following beneficial effects:

[0035] The voice data warehouse construction method, apparatus, storage medium, and electronic device provided in this application, after receiving voice data, extract business activity events from the voice data as target business activity events. A business activity event is an indivisible behavioral event; however, one business activity event may generate multiple requirements. That is, the analysis of voice data is based on business activity events, not requirements. The target business activity event is matched with a pre-generated first correspondence to obtain the corresponding business information domain as the target business information domain. The first correspondence is pre-generated and records the correspondence between business activity events and business information domains. The target business information domain is matched with a pre-generated second correspondence to obtain the corresponding dimension information as target dimension information. The second correspondence is the correspondence between business information domains and dimension information; that is, the target dimension information is also based on business activity events. The analysis objects corresponding to the target dimension information in the voice data are extracted as the information to be stored in the voice data warehouse, thus constructing the voice data warehouse. The above-mentioned solution of this application is based on business activity events in the entire process of voice data processing. Compared with the existing technology that takes the demand in voice data as the analysis object, the solution of this application determines the analysis object by business activity events, which makes it easier to classify voice data and make the classification results more reusable. Even when the amount of voice data is large, the analysis process of voice data can be mapped to the analysis process corresponding to the existing business abstract data in a timely manner, thereby improving the reusability of the analysis method and the flexibility of the voice data warehouse construction process. Attached Figure Description

[0036] Figure 1 This is a flowchart of a voice data warehouse construction method in one embodiment of this application;

[0037] Figure 2a A schematic diagram illustrating the correspondence between wake-up domains and business activity events in one embodiment of this application;

[0038] Figure 2bThis is a schematic diagram illustrating the correspondence between the natural language understanding & multiplexed arbitration domain and business activity events in one embodiment of this application;

[0039] Figure 3 This is a schematic diagram illustrating the correspondence between dimensional information and analysis objects in one embodiment of this application;

[0040] Figure 4 This is a structural block diagram of a voice data warehouse construction apparatus according to one embodiment of this application;

[0041] Figure 5 This is a schematic diagram of the hardware connections of an electronic device that performs the voice data warehouse construction method according to an embodiment of this application. Detailed Implementation

[0042] The specific embodiments of this application will be further described below with reference to the accompanying drawings.

[0043] It is readily understood that, based on the technical solution of this application, various structural and implementation methods can be interchanged by those skilled in the art without altering the essential spirit of this application. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this application and should not be considered as the entirety of this application or as limitations or restrictions on the technical solution of the application.

[0044] The directional terms such as up, down, left, right, front, back, front, back, top, and bottom mentioned or possibly used in this specification are defined relative to the structures shown in the accompanying drawings. They are relative concepts and may therefore vary depending on their location and usage. Therefore, these or other directional terms should not be interpreted as restrictive.

[0045] This application provides a method for constructing a voice data warehouse, applied in a system controller that processes voice data, such as... Figure 1 As shown, the method includes the following steps:

[0046] S10: Receive voice data and extract the business activity events in the voice data as target business activity events.

[0047] In this step, when voice data is received, it can be assumed that some electronic control units that process voice data have been activated. Therefore, the system controller already has the function of processing voice, such as obtaining voice data through voiceprint recognition and other methods.

[0048] Business activity events such as placing an order, making a payment, and getting a refund are all indivisible behavioral events; simply put, they are events that involve performing certain activities. Semantic analysis of voice data can identify the business activity events it contains. For example, if a user says "Pay the current refueling fee," the presence of the word "pay" indicates that the business activity event is payment.

[0049] S20: Match the target business activity event with a pre-generated first correspondence to obtain the business information domain corresponding to the target business activity event as the target business information domain; wherein, the first correspondence is the correspondence between the business activity event and the business information domain; wherein, each business activity event is an indivisible behavioral event, and each business information domain is a preset function extracted from historical business activity events.

[0050] As mentioned earlier, the business activity event is an indivisible behavioral event, typically including voice wake-up, voiceprint recognition, service expert, navigation intent, semantic slot filling, self-developed search, "see and speak" functionality, multi-turn navigation, multi-turn voice navigation, sending requests to services via voice, and receiving responses from services. The business information domain is information abstracted from the business activity event, usually representing preset functions. Different business activity events result in different business information domains. For example... Figure 2a and Figure 2b As shown, Figure 2a In this context, the business activity event involves waking up the vehicle, according to... Figure 2a The process involves the user first stating the vehicle's name, such as "XX classmate". The subsequent steps include various ways to wake up the vehicle, and regardless of the method, the vehicle will be in a running state. The vehicle will then broadcast a welcome message. The default function of the entire business activity event is to wake up the vehicle, and the corresponding business information domain is the wake-up domain. Figure 2b In this context, the business activity event needs to determine its semantic meaning. This is achieved by performing offline and online semantic recognition on the speech recognition results, then fusing the offline and online recognition results to obtain the first semantic recognition result. Combining this with the principle that "what is visible is what can be said" from the speech recognition results yields the second semantic recognition result. Finally, multi-path semantic arbitration is used to determine the final semantic recognition result. Therefore, its default function corresponds to natural language understanding and semantic multi-path arbitration; hence, the business information domain is the natural language understanding & multi-path arbitration domain. Typically, the business information domain in a vehicle can include a voice wake-up domain, a voiceprint domain, a vertical intent domain, a semantic slot filling domain, a self-developed large search domain, a free dialogue domain, a multi-turn dialogue domain, and a multi-modal data domain, etc.

[0051] By pre-generating a first correspondence to record the correspondence between known business activity events and business information domains, once the target business activity event is determined from the voice data, the target business information domain can be directly determined through the first correspondence.

[0052] S30: Match the target business information domain with the pre-generated second correspondence to obtain the dimension information corresponding to the target business information domain as the target dimension information; wherein, the second correspondence is the correspondence between the business information domain and the dimension information, and each of the dimension information is an analysis object extracted from the historical business information domain.

[0053] For each preset function's business information domain, the analysis objects to be identified are divided into multiple dimensions. A dimension is a measurement environment used to reflect a type of analysis object in business activity events; it can also be called an entity object, such as a vehicle dimension, voiceprint dimension, or date dimension. For example, when the business information domain is a wake-up domain, it implements a preset function to wake up a certain target. In this case, the object to be analyzed includes which target is being woken up. If the target is a vehicle, then the vehicle is one dimension. When waking up the video playback software in the vehicle, the video playback software can be considered a dimension. The analysis objects can be adjusted according to the specific application settings, and each analysis object can serve as one of the dimension information.

[0054] By pre-generating a second correspondence to record the correspondence between business information domains and dimension information, once the target information domain is determined, the target dimension information can be directly determined through the second correspondence.

[0055] S40: Extract the analysis object corresponding to the target dimension information from the speech data as the information to be stored in the speech data warehouse, and construct the speech data warehouse.

[0056] like Figure 3 As shown, the vehicle dimension information can include various IDs as analysis objects, the voiceprint dimension can include various voiceprint attribute information as analysis objects, and the date dimension can include the operation time recorded when different operations are performed. By analyzing the voice data, the analysis objects contained in the target dimension information can be determined and stored. For example, if the voice data contains the OTA version and vehicle record-id from the vehicle dimension, these two analysis objects can be extracted as the information to be stored.

[0057] After receiving voice data, business activity events are extracted from the voice data as target business activity events. A business activity event is an indivisible behavioral event; however, a single business activity event may generate multiple requirements. Therefore, the analysis of the voice data is based on business activity events, not requirements. The target business activity event is matched with a pre-generated first correspondence to obtain the corresponding business information domain, which is then used as the target business information domain. The first correspondence is pre-generated and records the correspondence between business activity events and business information domains. The target business information domain is then matched with a pre-generated second correspondence to obtain the corresponding dimension information, which is used as the target dimension information. The second correspondence is also based on business activity events. The analysis objects corresponding to the target dimension information are extracted from the voice data as the information to be stored in the voice data warehouse, thus constructing the voice data warehouse. The above-mentioned solution of this application is based on business activity events in the entire process of voice data processing. Compared with the existing technology that takes the demand in voice data as the analysis object, the solution of this application determines the analysis object by business activity events, which makes it easier to classify voice data and make the classification results more reusable. Even when the amount of voice data is large, the analysis process of voice data can be mapped to the analysis process corresponding to the existing business abstract data in a timely manner, thereby improving the reusability of the analysis method and the flexibility of the voice data warehouse construction process.

[0058] In some embodiments of this application, step S10 includes:

[0059] S101: Convert the voice data into text information.

[0060] This can be achieved using a speech-to-text function program segment.

[0061] S102: Input the text information into a preset semantic model to obtain the business activity event represented by the text information; the preset semantic model is trained by taking text information samples as input and the business activity events corresponding to the text information samples as output.

[0062] The training process can be implemented using existing deep learning algorithms, such as CNNs (Convolutional Neural Networks). During training, text information samples are used as input, and a corresponding business activity event label is assigned to each text information sample. By training the CNN with a sufficient number of samples, a pre-defined semantic model can be obtained. Subsequently, as long as the text information to be analyzed is obtained, it can be input into the pre-defined semantic model to directly obtain the actual business activity event corresponding to the text information.

[0063] S103: The business activity event represented by the text information is taken as the target business activity event.

[0064] That is, the business activity event corresponding to the text information is the target business activity event.

[0065] This application utilizes a pre-trained semantic model, combined with mature speech-to-text functionality, to efficiently and accurately obtain target business activity events contained in voice data.

[0066] Preferably, the first correspondence in the above scheme is obtained in the following way:

[0067] S201: Obtain application process running data for historical business activity events.

[0068] Assuming this method is applied to the vehicle's infotainment system, then the application process data refers to the runtime data of the application processes within the infotainment system. These application processes include programs closely related to various vehicle functions, such as applications controlling a specific electronic control unit, applications receiving certain types of data, and applications analyzing the safety of the driving conditions. The runtime of each application process reflects the time required for the infotainment system to perform certain functions. Therefore, the runtime data of the application processes can reflect information related to the business information domain.

[0069] S202: Extract the last running application process recorded in the application process running data as the target process, take the function to be implemented by the target process as the preset function, and take the preset function as the name of the business information domain.

[0070] For example Figure 2a The business information field shown indicates the final moment of vehicle wake-up, where the application process controlling vehicle wake-up starts and records the start time. Therefore, the application process that controls vehicle wake-up in the last step is the target process. That is, the default function is wake-up, and the business information field is the wake-up field.

[0071] S203: The correspondence between the historical business activity events and the business information domain is taken as the first correspondence.

[0072] Figure 2a Each business activity event shown in the diagram is mapped to a wake-up domain. Similarly, each historical business activity event is mapped to a business information domain and a corresponding relationship is obtained by following the steps described above. All the corresponding relationships are summarized into data tables or other forms, and the first corresponding relationship can be determined.

[0073] The proposed solution determines the business information domain based on the application process's runtime data, which has the advantages of standardization and normalization. Furthermore, the application process's runtime data is easy to obtain, thus improving data analysis efficiency.

[0074] Furthermore, in the above scheme, the step of extracting the last running application process recorded in the application process running data as the target process, and taking the function to be implemented by the target process as the preset function, includes: when the target process corresponds to the wake-up process of a vehicle or a set target, the preset function is wake-up, corresponding to the wake-up domain in the business information domain; when the target process corresponds to the voiceprint analysis process, the preset function is voiceprint recognition, corresponding to the voiceprint recognition domain in the business information domain; when the target process is the same as the previously running application processes in multiple consecutive steps, the preset function is multi-turn dialogue, corresponding to the multi-turn dialogue domain in the business information domain. In addition to determining the business information domain based on the function of the application process, this application scheme can also determine the business information domain based on the running relationship between the application process and the application processes in previous steps. This method is very necessary for some application scenarios. For example, when the business activity event corresponding to the voice data is "play a movie named A", the playback software displays multiple pages of search results, with each page of search results arranged according to relevance. At this point, it's necessary to continue determining whether the desired movie is available on the current page. If not, a next page needs to be selected, and the process continues after the next page is displayed. This business activity event remains unchanged, but multiple rounds of selection are required. Throughout this process, the application process continuously controls the playback software to execute step-by-step operations. This type of scenario can be categorized as a multi-turn dialogue domain. This application's solution enriches the rules for dividing business information domains, making the division more flexible and adaptable to a wider range of application scenarios.

[0075] Preferably, in the above scheme, the step of matching the target business information domain with a pre-generated second correspondence to obtain the dimension information corresponding to the target business information domain is taken as the target dimension information:

[0076] The dimension information in the second correspondence includes multi-level dimension information; wherein, the subsequent level dimension information is obtained by dividing the previous level dimension information according to a set classification method; the target dimension information includes multi-level target dimension information obtained by matching the target business information domain with the second correspondence. Taking the date dimension as an example, the date dimension can be used as a first-level dimension, but since the date includes year, month, and day, it can be further divided into second-level dimensions such as "year dimension," "month dimension," and "day dimension." Similarly, the geographical dimension can be used as a first-level dimension, and it can be further classified into second-level dimensions such as "country name dimension," "country ID dimension," and "province name dimension." In specific implementation, the number of dimension levels can be determined according to the actual situation.

[0077] The results of summarizing multiple dimensions can be stored using a matrix or data table. For example, a two-dimensional data table can be selected, with each row recording the first-level dimensions and each column recording the second-level dimensions. Through the above-described scheme of this application, the number of levels of dimensional information can be set as needed, thereby refining the analysis of voice data and resulting in more accurate final results.

[0078] like Figure 4 As shown in the figure, this application provides a voice data warehouse construction apparatus, including:

[0079] The voice data receiving module 10 is configured to receive voice data and extract business activity events from the voice data as target business activity events. Upon receiving voice data, it can be assumed that some electronic control units that process the voice data have been activated. Therefore, the system controller already possesses the function of voice processing; for example, the user's speech can be obtained through voiceprint recognition. Business activity events such as placing an order, making a payment, and issuing a refund are all indivisible behavioral events—in simpler terms, events that execute certain activities. Semantic analysis of the voice data can determine the business activity events it contains.

[0080] The first matching module 20 is configured to match the target business activity event with a pre-generated first correspondence to obtain a business information domain corresponding to the target business activity event as the target business information domain. The first correspondence is a relationship between business activity events and business information domains. Each business activity event is an indivisible behavioral event, and each business information domain is a preset function extracted from historical business activity events. The business information domain is information obtained by abstracting business activity events, typically a preset function. Different business activity events are abstracted to obtain different business information domains. By pre-generating the first correspondence to record the known correspondence between business activity events and business information domains, the target business information domain can be directly determined through the first correspondence after the target business activity event is determined from the voice data.

[0081] The second matching module 30 is configured to match the target business information domain with a pre-generated second correspondence to obtain the dimension information corresponding to the target business information domain as the target dimension information. The second correspondence is a relationship between business information domains and dimension information, where each dimension information is an analysis object extracted from a historical business information domain. For each preset function's business information domain, the analysis object to be determined is divided into multiple dimensions. A dimension is a measurement environment used to reflect a type of analysis object in business activity events; it can also be called an entity object, such as a vehicle dimension, voiceprint dimension, or date dimension. By pre-generating a second correspondence to record the correspondence between business information domains and dimension information, once the target information domain is determined, the target dimension information can be directly determined through the second correspondence.

[0082] The construction module 40 is configured to extract the analysis objects corresponding to the target dimension information from the speech data as the information to be stored in the speech data warehouse, and to construct the speech data warehouse. By analyzing the speech data, the analysis objects contained in the target dimension information can be identified and stored.

[0083] The device provided in this application determines the analysis object based on business activity events, making it easier to classify voice data and making the classification results more reusable. Even when the amount of voice data is large, the analysis process of voice data can be mapped to the analysis process corresponding to existing business abstract data in a timely manner, thereby improving the reusability of the analysis method and the flexibility of the voice data warehouse construction process.

[0084] Furthermore, the construction module 40 is located in the data application layer. The data application layer can directly obtain relevant operational data for application scenarios, facilitating the extraction of relevant information for analyzing business activity events. The data application layer has the advantage of analyzing business activity events.

[0085] This application provides a computer program product in some embodiments, including a computer program / instructions, which, when executed by a processor, implement the steps of the voice data warehouse construction method described in any of the above embodiments.

[0086] In some embodiments of this application, a computer-readable storage medium is provided, wherein program information is stored in the storage medium, and a computer reads the program information and executes the voice data warehouse construction method described in any of the above method embodiments.

[0087] This application also provides an electronic device, such as... Figure 5As shown, the electronic device includes at least one processor 51 and at least one memory 52. ​​The at least one memory 52 stores program information. After reading the program information, the at least one processor 51 executes the voice data warehouse construction method described in any of the above method embodiments. The device may further include an input device 53 and an output device 54. The processor 51, memory 52, input device 53, and output device 54 can be communicatively connected. The memory 52, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules. The processor 51 executes various functional applications and data processing by running the non-volatile software programs, instructions, and modules stored in the memory 52, thereby implementing the voice data warehouse construction method provided in any of the above embodiments. The memory 52 may include a program storage area and a data storage area. The program storage area may store the operating system and at least one application program required for a function; the data storage area may store data created according to the use of the voice data warehouse construction method, etc. Furthermore, memory 52 may include high-speed random access memory and non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, memory 52 may optionally include memory remotely located relative to processor 51, and these remote memories may be connected via a network to the apparatus performing the voice data warehouse construction method. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof. Input device 53 may receive user clicks and generate signal inputs related to user settings and function control of the voice data warehouse construction method. Output device 54 may include a display device such as a display screen. When the one or more modules are stored in memory 52 and are run by the one or more processors 51, the voice data warehouse construction method in any of the above method embodiments is executed.

[0088] As needed, the above technical solutions can be combined to achieve the best technical effect.

[0089] The above are merely the principles and preferred embodiments of this application. It should be noted that, for those skilled in the art, several other modifications can be made based on the principles of this application, and these modifications should also be considered within the scope of protection of this application.

Claims

1. A method for constructing a voice data warehouse, characterized in that, include: Receive voice data and extract the business activity events in the voice data as target business activity events; The target business activity event is matched with a pre-generated first correspondence to obtain a business information domain corresponding to the target business activity event as the target business information domain; wherein, the first correspondence is the correspondence between the business activity event and the business information domain; the first correspondence is obtained in the following way: obtaining application process running data of historical business activity events; extracting the last running application process recorded in the application process running data as the target process, taking the function to be implemented by the target process as the preset function, and taking the preset function as the name of the business information domain; taking the correspondence between the historical business activity event and the business information domain as the first correspondence; wherein, each business activity event is an indivisible behavioral event, and each business information domain is a preset function extracted from the historical business activity event, including: when the target process is the same as the previously running consecutive multi-step application process, the preset function is multi-turn dialogue, corresponding to the business information domain being a multi-turn dialogue domain; The target business information domain is matched with a pre-generated second correspondence to obtain the dimension information corresponding to the target business information domain as the target dimension information; wherein, the second correspondence is the correspondence between the business information domain and the dimension information, and each of the dimension information is an analysis object extracted from the historical business information domain; Extract the analysis objects corresponding to the target dimension information from the speech data as the information to be stored in the speech data warehouse, and construct the speech data warehouse.

2. The method for constructing a voice data warehouse according to claim 1, characterized in that, The step of receiving voice data and extracting business activity events from the voice data as target business activity events includes: Convert the voice data into text information; The text information is input into a preset semantic model to obtain the business activity event represented by the text information; the preset semantic model is trained by taking text information samples as input and the business activity events corresponding to the text information samples as output. The business activity event represented by the text information is taken as the target business activity event.

3. The method for constructing a voice data warehouse according to claim 1, characterized in that, The step involves matching the target business information domain with a pre-generated second correspondence to obtain the dimension information corresponding to the target business information domain, which is then used as the target dimension information. The dimensional information in the second correspondence includes multi-level dimensional information; wherein, the dimensional information at the next level is obtained by dividing the dimensional information at the previous level according to a set classification method; The target dimension information includes multi-level target dimension information obtained by matching the target business information domain with the second correspondence.

4. The method for constructing a voice data warehouse according to claim 3, characterized in that, The step of extracting the analysis object corresponding to the target dimension information from the speech data as the information to be stored in the speech data warehouse includes: The analysis objects corresponding to each level of target dimension information in the voice data are divided into a group of information to be stored. The different groups of information to be stored are stored in different storage areas of the voice data warehouse.

5. A voice data warehouse construction device, characterized in that, include: The voice data receiving module is configured to receive voice data and extract business activity events from the voice data as target business activity events; The first matching module is configured to match the target business activity event with a pre-generated first correspondence to obtain a business information domain corresponding to the target business activity event as the target business information domain; wherein, the first correspondence is the correspondence between the business activity event and the business information domain; the first correspondence is obtained by: acquiring application process running data of historical business activity events; extracting the last running application process recorded in the application process running data as the target process, taking the function to be implemented by the target process as the preset function, and taking the preset function as the name of the business information domain; taking the correspondence between the historical business activity event and the business information domain as the first correspondence; wherein, each business activity event is an indivisible behavioral event, and each business information domain is a preset function extracted from the historical business activity event, including: when the target process is the same as a previously running consecutive multi-step application process, the preset function is a multi-turn dialogue, corresponding to the business information domain being a multi-turn dialogue domain; The second matching module is configured to match the target business information domain with a pre-generated second correspondence relationship to obtain the dimension information corresponding to the target business information domain as the target dimension information; wherein, the second correspondence relationship is the correspondence relationship between the business information domain and the dimension information, and each of the dimension information is an analysis object extracted from the historical business information domain; The construction module is configured to extract the analysis objects corresponding to the target dimension information in the speech data as the information to be stored in the speech data warehouse, and to construct the speech data warehouse.

6. The voice data warehouse construction apparatus according to claim 5, characterized in that: The building module is located in the data application layer of the operating system.

7. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the voice data warehouse construction method according to any one of claims 1-4.

8. A computer-readable storage medium, characterized in that, The storage medium stores program information, and after the computer reads the program information, it executes the voice data warehouse construction method according to any one of claims 1-4.

9. An electronic device, characterized in that, The electronic device includes at least one processor and at least one memory, wherein at least one memory stores program information, and at least one processor reads the program information and executes the voice data warehouse construction method according to any one of claims 1-4.

Images