Architectural knowledge question-answering system and architectural knowledge answer information generation method

Abstract

The embodiment of the present disclosure discloses an architectural knowledge question-and-answer system and a method for generating architectural knowledge answer information. A specific implementation of the system includes: a client, a server, and a database server, wherein: the database server is configured to create an architectural knowledge base and an architectural knowledge graph; the client is configured to send architectural knowledge demand information to the server; the server is configured to perform: extracting deep semantic information corresponding to the architectural knowledge demand information; the database server is further configured to retrieve the created architectural knowledge base and architectural knowledge graph; the server is further configured to perform the following steps: inputting architectural knowledge demand information, intent recognition information, and architectural knowledge base query information into an architectural knowledge question-and-answer model; determining an entity concept information set and initial architectural knowledge answer information as architectural knowledge answer information; and sending architectural knowledge answer information to the client. This implementation improves the accuracy of architectural knowledge answer information.

Description

Technical Field

[0001] The embodiments of the present disclosure relate to the field of computer technology, and in particular to an architectural knowledge question-answering system and an architectural knowledge answer information generating method. Background Art

[0002] With the continuous advancement of artificial intelligence technology, question-answering systems have been widely used in all aspects of life. The architectural knowledge question-answering system is a system that provides architectural knowledge question-answering services to users and generates architectural knowledge answer information corresponding to architectural knowledge demand information. At present, when generating architectural knowledge answer information, the commonly used method is to directly generate architectural knowledge answer information based on architectural knowledge demand information through a pre-trained question-answering model.

[0003] However, when using the above method to generate architectural knowledge response information, the following technical problems often occur:

[0004] First, directly based on the architectural knowledge demand information, the architectural knowledge reply information is generated through the pre-trained question-answering model. The question-answering model may be limited to the information contained in the training data. If the training data used to train the question-answering model contains less architectural knowledge information, the accuracy of the architectural knowledge reply information will be low. At the same time, the training of the question-answering model is based on historical data sets. The architectural knowledge contained in the historical data sets is static, which may cause the architectural knowledge reply information input by the question-answering model to fail to reflect the emerging architectural technology or architectural knowledge, and thus, the real-time performance of the architectural knowledge reply information is poor.

[0005] Continuing, in the process of adopting technical solutions to solve the above-mentioned technical problem 1, the following technical problems often occur: before generating architectural knowledge reply information, it is necessary to retrieve architectural knowledge base query information corresponding to architectural knowledge demand information. The conventional solution for retrieving architectural knowledge base query information corresponding to architectural knowledge demand information is generally: directly retrieve each keyword in the architectural knowledge demand information through a preset database, and obtain each retrieval information corresponding to each keyword as architectural knowledge base query information. However, the above conventional solution directly retrieves each keyword in the architectural knowledge demand information through a preset database, and obtains each retrieval information corresponding to each keyword as architectural knowledge base query information, but still has the following problems:

[0006] Second, directly searching for each keyword in the architectural knowledge demand information through the preset database and obtaining each search information corresponding to each keyword as the architectural knowledge base query information may result in poor correlation between the various search information obtained. The search information with poor correlation may contain misleading information, thereby resulting in low accuracy of the architectural knowledge response information generated based on the architectural knowledge base query information.

[0007] The above information disclosed in this Background section is only for enhancement of understanding of the background of the inventive concept and therefore it may contain information that does not form the prior art that is already known in this country to a person of ordinary skill in the art. Summary of the invention

[0008] The content of this disclosure is used to introduce concepts in a brief form, which will be described in detail in the detailed implementation section below. The content of this disclosure is not intended to identify the key features or essential features of the technical solution claimed for protection, nor is it intended to limit the scope of the technical solution claimed for protection.

[0009] Some embodiments of the present disclosure propose an architectural knowledge question-answering system and an architectural knowledge answer information generation method to solve one or more of the technical intentions mentioned in the above background technology section.

[0010] In a first aspect, some embodiments of the present disclosure provide a building knowledge question-and-answer system, the system comprising: a client, a server, and a database server, wherein: the database server is configured to create a building knowledge base and a building knowledge graph; the client is communicatively connected to the server, wherein the client is configured to send building knowledge demand information to the server; the server is configured to perform the following steps: in response to receiving the building knowledge demand information sent by the client, extracting deep semantic information corresponding to the building knowledge demand information, wherein the deep semantic information includes: intent recognition information and a keyword set; sending the keyword set to the database server; the database server is communicatively connected to the server, wherein the database server is further configured to retrieve the keyword set in response to receiving the keyword set The created architectural knowledge base and architectural knowledge graph are used to generate a query information set corresponding to the above-mentioned keyword set, wherein each query information in the above-mentioned query information set includes: an entity concept information set and architectural knowledge base query information; the above-mentioned server is further configured to perform the following steps: in response to receiving the query information set sent by the above-mentioned database server, the above-mentioned architectural knowledge demand information, the above-mentioned intention recognition information and the architectural knowledge base query information included in the above-mentioned query information set are input into a pre-trained architectural knowledge question and answer model to obtain initial architectural knowledge reply information corresponding to the above-mentioned architectural knowledge demand information; the entity concept information set included in the above-mentioned query information set and the above-mentioned initial architectural knowledge reply information are determined as architectural knowledge reply information; the above-mentioned architectural knowledge reply information is sent to the above-mentioned client, so that the above-mentioned client displays the above-mentioned architectural knowledge reply information.

[0011] In a second aspect, some embodiments of the present disclosure provide a method for generating architectural knowledge reply information, the method comprising: in response to receiving the architectural knowledge demand information sent by the client, extracting deep semantic information corresponding to the architectural knowledge demand information, wherein the deep semantic information comprises: intent recognition information and a keyword set; sending the keyword set to the database server so that the database server can retrieve the created architectural knowledge base and architectural knowledge graph to generate a query information set corresponding to the keyword set; in response to receiving the query information set sent by the database server, inputting the architectural knowledge demand information, the intent recognition information and the architectural knowledge base query information included in the query information set into a pre-trained architectural knowledge question and answer model to obtain initial architectural knowledge reply information corresponding to the architectural knowledge demand information; determining the entity concept information set included in the query information set and the initial architectural knowledge reply information as architectural knowledge reply information; sending the architectural knowledge reply information to the client so that the client can display the architectural knowledge reply information.

[0012] The above-mentioned embodiments of the present disclosure have the following beneficial effects: through the architectural knowledge question-and-answer system of some embodiments of the present disclosure, the real-time and accuracy of architectural knowledge reply information are improved. Specifically, the reason for the poor real-time and accuracy of architectural knowledge reply information is that: directly based on the architectural knowledge demand information, the architectural knowledge reply information is generated by a pre-trained question-and-answer model, and the question-and-answer model may be limited to the information contained in the training data. If the training data used to train the question-and-answer model contains less architectural knowledge information, the accuracy of the architectural knowledge reply information will be low. At the same time, the training of the question-and-answer model of the question-and-answer model is based on the historical data set. The architectural knowledge contained in the historical data set is static, which causes the architectural knowledge reply information input by the question-and-answer model to be unable to reflect the newly emerging architectural technology or architectural knowledge, and then, the real-time of the architectural knowledge reply information is poor. Based on this, the architectural knowledge question-and-answer system of some embodiments of the present disclosure includes: a client, a server, and a database server, wherein: first, the database server is configured to create an architectural knowledge base and an architectural knowledge graph. Then, the client is connected to the server in communication, wherein the client is configured to send the architectural knowledge demand information to the server. Thus, the architectural knowledge demand information can be sent to the server. Next, the server is configured to perform the following steps: in response to receiving the architectural knowledge demand information sent by the client, extracting the deep semantic information corresponding to the architectural knowledge demand information, wherein the deep semantic information includes: intent recognition information and a keyword set. Thus, the intent recognition information for generating initial architectural knowledge reply information can be obtained. Then, the keyword set is sent to the database server. Next, the database server is communicatively connected to the server, wherein the database server is further configured to retrieve the created architectural knowledge base and architectural knowledge graph in response to receiving the keyword set, so as to generate a query information set corresponding to the keyword set, wherein each query information in the query information set includes: an entity concept information set and architectural knowledge base query information. Thus, an entity concept information set for generating architectural knowledge reply information and architectural knowledge base query information with good real-time performance can be obtained. Next, the server is further configured to perform the following steps: Then, in response to receiving the query information set sent by the database server, the architectural knowledge requirement information, the intention recognition information and the architectural knowledge base query information included in the query information set are input into a pre-trained architectural knowledge question-answering model to obtain initial architectural knowledge answer information corresponding to the architectural knowledge requirement information. Thus, the architectural knowledge question-answering model that generates answer information for architectural knowledge questions can generate more accurate initial architectural knowledge answer information based on the architectural knowledge requirement information, the intention recognition information and the architectural knowledge base query information with better real-time performance.Afterwards, the entity concept information set included in the query information set and the initial architectural knowledge answer information are determined as architectural knowledge answer information. Thus, architectural knowledge answer information with richer answer information can be obtained. Finally, the architectural knowledge answer information is sent to the client so that the client can display the architectural knowledge answer information. Thus, the architectural knowledge answer information can be displayed through the client. Also, because the architectural knowledge base and architectural knowledge graph created by retrieving the database server are adopted, architectural knowledge base query information with good real-time performance and corresponding to the architectural knowledge demand information is generated. On the basis of the architectural knowledge demand information, the intention recognition information and the architectural knowledge base query information with good real-time performance, the architectural knowledge question-answering model that generates answer information for architectural knowledge questions generates initial architectural knowledge answer information with higher accuracy and better real-time performance. The architectural knowledge answer information is generated on the basis of the initial architectural knowledge answer information with higher accuracy and better real-time performance. The accuracy and real-time performance of the architectural knowledge answer information are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the accompanying drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and that components and elements are not necessarily drawn to scale.

[0014] Figure 1 is an architecture diagram of an exemplary system of the architectural knowledge question answering system according to the present disclosure;

[0015] Figure 2 Flow chart of some embodiments of the method for generating architectural knowledge answer information according to the present disclosure. DETAILED DESCRIPTION

[0016] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments set forth herein. On the contrary, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.

[0017] It should also be noted that, for ease of description, only the parts related to the invention are shown in the drawings. In the absence of conflict, the embodiments and features in the embodiments of the present disclosure can be combined with each other.

[0018] It should be noted that the concepts such as "first" and "second" mentioned in the present disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0019] It should be noted that the modifications of "one" and "plurality" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0020] The names of the messages or information exchanged between multiple devices in the embodiments of the present disclosure are only used for illustrative purposes and are not used to limit the scope of these messages or information.

[0021] The present disclosure will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

[0022] Figure 1 An exemplary system architecture 100 of a heterogeneous data exchange system is shown to which some embodiments of the present disclosure may be applied.

[0023] like Figure 1 As shown, the system architecture 100 may include: a client 101, a server 102, a database server 103, a network 104, and a network 105. The network 104 is used as a medium for providing a communication link between the client 101 and the server 102. The network 105 is used as a medium for providing a communication link between the server 102 and the database server 103. The networks 104 and 105 may include various connection types, such as wired, wireless communication links, or optical fiber cables, etc.

[0024] In some embodiments, the database server 103 may be configured to create a building knowledge base and a building knowledge graph.

[0025] In some optional implementations of some embodiments, the database server 103 may create a building knowledge base and a building knowledge graph through the following steps:

[0026] The first step is to obtain a building knowledge information set from a preset network storage device. The preset network storage device may be one of the following: a network storage device, a cloud server. The building knowledge information in the building knowledge information set may include text information related to architecture, engineering, building design, construction technology, and building materials. For example, the building knowledge information may be "Building A was designed by architect B".

[0027] The second step is to create an initial database according to the preset demand information. The preset demand information may indicate the specific requirements for creating a database. The preset demand information may include but is not limited to at least one of the following: database usage scenario information and database structure information. The database usage scenario information may indicate the situation in which the database is used to store and manage data under the preset scenario. The database structure information may indicate the structure of the database to be created. For example, the database usage scenario information may be "storage and retrieval of architectural knowledge information". The database structure information may be "relational database structure". In practice, the database server 103 may create a database corresponding to the preset demand information as the initial database through a database management system.

[0028] The third step is to store the building knowledge information set into the initial database to update the initial database.

[0029] In the fourth step, the updated initial database is determined as the architectural knowledge base.

[0030] Step 5: Entity recognition is performed on the above-mentioned architectural knowledge information set to obtain an entity information set. As an example, the above-mentioned architectural knowledge information set may be {"Building A was designed by Architect B", "Architect B graduated from University D", "University D is located to the south of Building A"}. One entity information in the above-mentioned entity information set may be an entity name. For example, the above-mentioned entity information set may be {Building A, Architect B, University D}.

[0031] The sixth step is to determine each entity information in the above entity information set as each graph node.

[0032] The seventh step is to perform entity relationship extraction on the above-mentioned architectural knowledge information set based on the above-mentioned entity information to obtain the entity relationship information. Among them, each entity relationship information in the above-mentioned entity relationship information includes: starting entity information, relationship information, and pointing entity information. In practice, for each architectural knowledge information in the above-mentioned architectural knowledge information set, the above-mentioned database server 103 can perform entity relationship extraction on the above-mentioned architectural knowledge information to obtain entity relationship information. As an example, the above-mentioned architectural knowledge information can be "Building A is designed by Architect B", and the above-mentioned entity relationship information can be "Building A-Designed by XXX-Architect B". The above-mentioned starting entity information can be "Building A". The above-mentioned relationship information can be "Designed by XXX". The above-mentioned pointing entity information can be "Architect B".

[0033] Step 8: For each entity relationship information in the above entity relationship information, perform the following steps:

[0034] In the first sub-step, each graph node corresponding to the starting point entity information and the pointing entity information included in the above-mentioned entity relationship information in the above-mentioned graph nodes is determined as a graph node group.

[0035] The second sub-step is to determine the relationship information included in the above-mentioned entity relationship information as the edge information between the graph nodes in the above-mentioned graph node group.

[0036] The ninth step is to generate an initial architectural knowledge graph based on each graph node group of the determined edge information. In practice, the database server 103 may call the first preset API interface to create a knowledge graph corresponding to each graph node group of the determined edge information as the initial architectural knowledge graph. The first preset API interface may be used to create a knowledge graph based on each node of the determined edge information.

[0037] The tenth step is to generate an architectural knowledge graph based on the above entity information and the initial architectural knowledge graph.

[0038] In some optional implementations of some embodiments, the database server 103 may generate a building knowledge graph based on the above-mentioned entity information and the initial building knowledge graph through the following steps:

[0039] In the first step, for each of the above entity information, perform the following steps:

[0040] The first sub-step is to obtain entity concept information corresponding to the entity information from the preset network storage device, wherein the entity concept information may be text information representing the entity concept.

[0041] In the second sub-step, the graph node corresponding to the above entity information in the above initial architectural knowledge graph is determined as the target graph node.

[0042] The third sub-step is to add the entity concept information as the graph node attribute information to the target graph node to update the initial architectural knowledge graph. In practice, the database server 103 can add the entity concept information as the graph node attribute information to the target graph node through the second preset API interface. The second preset API interface can add node attribute information to the graph node in the knowledge graph.

[0043] In the second step, the updated initial architectural knowledge graph is determined as the architectural knowledge graph.

[0044] In some embodiments, the client 101 may be configured to send the architectural knowledge demand information to the server. The client 101 may be a device for receiving architectural knowledge questions input by a user. For example, the device may be, but is not limited to, one of the following: a smart phone, a computer. The architectural knowledge demand information may be text information indicating architectural knowledge questions.

[0045] In some embodiments, the server 102 may be configured to perform the following steps:

[0046] First, in response to receiving the architectural knowledge requirement information sent by the client, extracting deep semantic information corresponding to the architectural knowledge requirement information, wherein the deep semantic information includes: intention recognition information and a keyword set.

[0047] In some optional implementations of some embodiments, the server 102 may extract the deep semantic information corresponding to the building knowledge requirement information in response to receiving the building knowledge requirement information sent by the client through the following steps:

[0048] In the first step, the architectural knowledge demand information is input into a pre-trained keyword extraction model to obtain a keyword set corresponding to the architectural knowledge demand information. The keyword extraction model may be a keyword extraction model that takes the architectural knowledge demand information as input and the keyword set as output.

[0049] In the second step, the above-mentioned building knowledge demand information is input into a pre-trained intention recognition model to obtain intention recognition information corresponding to the above-mentioned building knowledge demand information.

[0050] The third step is to determine the above keyword set and the above intent recognition information as deep semantic information.

[0051] In some optional implementations of some embodiments, the server 102 may input the building knowledge requirement information into a pre-trained intention recognition model through the following steps to obtain intention recognition information corresponding to the building knowledge requirement information:

[0052] The first step is to input the above-mentioned building knowledge demand information into the input layer of the above-mentioned intention recognition model to obtain the word segmentation sequence corresponding to the above-mentioned building knowledge demand information. Among them, the above-mentioned intention recognition model includes the above-mentioned input layer, word embedding layer, relative position encoding layer, stacked attention module, and intention classification layer. The above-mentioned stacked attention module includes various intention extraction modules, and each of the above-mentioned intention extraction modules includes a self-attention layer and a feedforward neural network layer. The above-mentioned input layer can take the building knowledge demand information as input and the word segmentation sequence as input. The above-mentioned word segmentation sequence can be a word sequence obtained by performing word segmentation processing on the above-mentioned building knowledge prelude information. For example, the above-mentioned building knowledge demand information can be "I want to know how green buildings can apply new technologies to improve the comfort of living environment."

[0053] The second step is to input the above word segmentation sequence into the above word embedding layer to obtain a word vector sequence corresponding to the above word segmentation sequence. A word vector in the above word vector sequence can represent a word in the above word segmentation sequence.

[0054] In the third step, the word vector sequence is input into the relative position encoding layer to obtain a relative position embedding vector sequence. The relative position encoding layer may be a position encoding layer. The relative position embedding vector sequence may be a word vector sequence with relative position encoding.

[0055] The fourth step is to input the relative position embedding vector sequence into the stacked attention module to obtain a weighted transformation word vector sequence corresponding to the relative position embedding vector sequence. The weighted transformation word vector sequence may be a vector sequence obtained by performing nonlinear transformation and weighting processing on each relative position embedding vector in the relative position embedding vector sequence.

[0056] The fifth step is to input the above-mentioned weighted transformation word vector sequence into the above-mentioned intention classification layer to obtain the intention recognition information corresponding to the above-mentioned weighted transformation word vector sequence. Among them, the above-mentioned intention classification layer can be a classification layer that takes the weighted transformation word vector sequence as input information and takes the intention recognition information as output information. For example, the above-mentioned intention recognition information can be "understand the application of green building technology." As an example, the first sample set can be obtained first. Among them, the first sample in the above-mentioned first sample set includes sample building knowledge requirement information and sample intention recognition information corresponding to the sample building knowledge requirement information. Then, the sample building knowledge requirement information included in each sample in the first sample set is used as input, and the sample intention recognition information included in the sample is used as the expected output, and the intention recognition model is trained.

[0057] Second, the keyword set is sent to the database server.

[0058] In some embodiments, the database server 103 may be further configured to retrieve the created architectural knowledge base and architectural knowledge graph in response to receiving the keyword set to generate a query information set corresponding to the keyword set. The query information in the query information set includes: entity concept information set and architectural knowledge base query information.

[0059] In some optional implementations of some embodiments, the database server 103 may retrieve the created architectural knowledge base and architectural knowledge graph through the following steps to generate a query information set corresponding to the keyword set:

[0060] The first step is to retrieve the entity concept information corresponding to each keyword in the above keyword set from the above architectural knowledge graph.

[0061] The second step is to determine the retrieved entity concept information as an entity concept information set.

[0062] The third step is to determine the above entity concept information set as query information.

[0063] The fourth step is to traverse each graph node corresponding to the above keyword set and the edge information between the above graph nodes from the above architectural knowledge graph to obtain each entity relationship information corresponding to the above keyword set.

[0064] In the fifth step, each entity relationship information corresponding to the above keyword set is determined as each entity relationship information to be retrieved.

[0065] Step 6: for each of the entity relationship information to be retrieved, retrieve the storage document corresponding to the entity relationship information to be retrieved from the architectural knowledge base, wherein the storage content in the storage document includes text information.

[0066] The seventh step is to determine the retrieved storage documents as candidate storage documents.

[0067] Step 8: For each candidate storage document in the above candidate storage documents, perform the following steps:

[0068] The first sub-step is to determine the text information in the candidate storage document as the text information to be processed.

[0069] The second sub-step is to perform word segmentation on the above text information to be processed to obtain text phrases.

[0070] The third sub-step is to perform the following steps for each text word in the above text phrase:

[0071] Sub-step 1: determining a word frequency value of the text word in the text phrase, wherein the word frequency value may be a TF value (word frequency) of the text word in the text phrase.

[0072] Sub-step two: determining the above text words and the above word frequency values ​​as word frequency information.

[0073] The fourth sub-step is to perform deduplication processing on each determined word frequency information to obtain each deduplication word frequency information after deduplication processing as a deduplication word frequency information group. The deduplication word frequency information group corresponds to the candidate storage document.

[0074] The ninth step is to update each obtained deduplication word frequency information group to obtain each updated deduplication word frequency information group. In practice, the above-mentioned database server 103 can perform the following steps for each deduplication word frequency information in each obtained deduplication word frequency information group: For each deduplication word frequency information in the above-mentioned deduplication word frequency information group, perform the following generation steps: Determine the text word in the above-mentioned deduplication word frequency information as the target text word. Determine the reverse document frequency value of the above-mentioned target text word in each of the above-mentioned candidate storage documents. Add the above-mentioned reverse document frequency value to the above-mentioned deduplication word frequency information to update the above-mentioned deduplication word frequency information. Determine the deduplication word frequency information group after each deduplication word frequency information is updated as each updated deduplication word frequency information group.

[0075] The tenth step is to generate respective importance index vectors corresponding to the respective updated deduplication word frequency information groups based on the respective updated deduplication word frequency information groups. Among them, each importance index vector in the respective importance index vectors can be a vector representing the updated deduplication word frequency information. In practice, the database server 103 can convert the respective updated deduplication word frequency information groups (text information format) into respective importance index vectors (vector format) through data format conversion technology.

[0076] Step 11: Based on the keyword set, generate a keyword set search vector corresponding to the keyword set. In practice, first, the database server 103 can generate a TF-IDF vector corresponding to the keyword set through a TF-IDF algorithm. Then, the database server 103 can determine the TF-IDF vector as the keyword set search vector.

[0077] Step 12, based on the above-mentioned various importance index vectors and the above-mentioned keyword set retrieval vector, the candidate storage documents corresponding to the above-mentioned keyword set are screened out from the above-mentioned various candidate storage documents as screening storage documents. In practice, first, the above-mentioned database server 103 can determine the similarities between the above-mentioned various importance index vectors and the above-mentioned keyword set retrieval vector as various target similarities. Then, the above-mentioned database server 103 can determine the target similarity with the largest similarity value among the above-mentioned various target similarities as the screening similarity. Next, the above-mentioned database server 103 can determine the importance index vector corresponding to the above-mentioned screening similarity as the screening importance index vector. Finally, the above-mentioned database server 103 can determine the candidate storage documents corresponding to the above-mentioned screening importance index vector from the above-mentioned various candidate storage documents as screening storage documents. Among them, the above-mentioned similarity can be cosine similarity.

[0078] In the thirteenth step, the text information in the above-mentioned filtered storage document is determined as the architectural knowledge base query information as the query information.

[0079] In the fourteenth step, each of the determined query information is determined as a query information set.

[0080] The above technical solution and its related contents, as an inventive point of the embodiment of the present disclosure, solve the second technical problem mentioned in the background technology, "directly searching for each keyword in the architectural knowledge demand information through the preset database, obtaining each retrieval information corresponding to each keyword as the architectural knowledge base query information, the correlation between each retrieval information obtained may be poor, and the retrieval information with poor correlation may contain misleading information, thereby resulting in low accuracy of the architectural knowledge reply information generated based on the architectural knowledge base query information". The factors that lead to low accuracy of the generated architectural knowledge reply information are often as follows: directly searching for each keyword in the architectural knowledge demand information through the preset database, obtaining each retrieval information corresponding to each keyword as the architectural knowledge base query information, the correlation between each retrieval information obtained may be poor, and the retrieval information with poor correlation may contain misleading information, thereby resulting in low accuracy of the architectural knowledge reply information generated based on the architectural knowledge base query information. If the above factors are solved, the effect of improving the accuracy of the generated architectural knowledge reply information can be achieved. In order to achieve this effect, first, the entity concept information corresponding to each keyword in the above keyword set is retrieved from the above architectural knowledge graph. Then, the retrieved entity concept information is determined as an entity concept information set. Thus, the entity concept information set used to generate the query information can be obtained. Next, the entity concept information set is determined as query information. Then, each graph node corresponding to the keyword set and the edge information between the graph nodes are traversed from the architectural knowledge graph to obtain each entity relationship information corresponding to the keyword set. Then, each entity relationship information corresponding to the keyword set is determined as each entity relationship information to be retrieved. Then, for each of the entity relationship information to be retrieved, the storage document corresponding to the entity relationship information to be retrieved is retrieved from the architectural knowledge base, wherein the storage content in the storage document includes text information. Thus, each storage document used to determine the screening storage document can be obtained. Then, each retrieved storage document is determined as each candidate storage document. Then, for each of the candidate storage documents, the following steps are performed: In the first sub-step, the text information in the candidate storage document is determined as the text information to be processed. Thus, the text information to be processed for generating text phrases can be obtained. In the second sub-step, the text information to be processed is segmented to obtain the text phrases. The third sub-step is to perform the following steps for each text word in the above text phrase: Sub-step 1 is to determine the word frequency value of the above text word in the above text phrase. Sub-step 2 is to determine the above text word and the above word frequency value as word frequency information. The fourth sub-step is to perform deduplication processing on each determined word frequency information, and obtain each deduplication word frequency information after deduplication processing as a deduplication word frequency information group, wherein the deduplication word frequency information group corresponds to the above candidate storage document.Thus, the deduplication frequency information group of the candidate storage document can be obtained. Afterwards, the obtained deduplication frequency information groups are updated to obtain the updated deduplication frequency information groups. Thus, the deduplication frequency information group can be updated. Then, based on the updated deduplication frequency information groups, the importance index vectors corresponding to the updated deduplication frequency information groups are generated. Thus, the importance index vectors representing the importance of each candidate storage document can be generated. Then, based on the keyword set, the keyword set retrieval vector corresponding to the keyword set is generated. Thus, the keyword set retrieval vector corresponding to the keyword set of the building knowledge demand information can be obtained. Then, based on the importance index vectors and the keyword set retrieval vector, the candidate storage documents corresponding to the keyword set are selected from the candidate storage documents as the selected storage documents. Thus, the candidate storage documents related to each keyword in the keyword set are selected as the selected storage documents. Then, the text information in the selected storage document is determined as the architectural knowledge base query information as the query information. Thus, the text information in the storage document related to each keyword in the keyword set can be determined as the architectural knowledge base query information as the query information. Finally, the determined query information is determined as a query information set. Also, through each importance index vector and the above keyword set retrieval vector, candidate storage documents related to each keyword in the keyword set are screened out from each candidate storage document as screening storage documents. The text information in the storage document related to each keyword in the keyword set is determined as the architectural knowledge base query information as the query information, which reduces the possibility of misleading information in the architectural knowledge base query information, thereby improving the accuracy of the architectural knowledge reply information generated based on the architectural knowledge base query information.

[0081] In some embodiments, the server 102 may be further configured to perform the following steps:

[0082] First, in response to receiving the query information set sent by the above-mentioned database server, the above-mentioned architectural knowledge requirement information, the above-mentioned intention recognition information and the architectural knowledge base query information included in the above-mentioned query information set are input into a pre-trained architectural knowledge question and answer model to obtain initial architectural knowledge response information corresponding to the above-mentioned architectural knowledge requirement information.

[0083] In some optional implementations of some embodiments, the server 102 may input the architectural knowledge demand information, the intention recognition information, and the architectural knowledge base query information included in the query information set into a pre-trained architectural knowledge question-answering model through the following steps to obtain initial architectural knowledge answer information corresponding to the architectural knowledge demand information. The architectural knowledge question-answering model includes: a question-answering model, and a reply information optimization module:

[0084] In the first step, the above-mentioned building knowledge requirement information is input into the above-mentioned question-answering model to obtain the initial reply information corresponding to the above-mentioned building knowledge requirement information. As an example, the second sample set can be first obtained. Among them, the second sample in the above-mentioned second sample set includes sample building knowledge requirement information, intention recognition information, building knowledge base query information and sample initial building knowledge reply information corresponding to the sample building knowledge requirement information, intention recognition information and building knowledge base query information. Then, the sample building knowledge requirement information, intention recognition information and building knowledge base query information included in each sample in the second sample set are used as input, and the sample initial building knowledge reply information included in the sample is used as the expected output, and the building knowledge question-answering model is trained.

[0085] In the second step, the above-mentioned initial reply information, the above-mentioned intention recognition information and the above-mentioned architectural knowledge base query information are input into the encoder layer of the above-mentioned reply information optimization module to obtain a reply information vector corresponding to the above-mentioned initial reply information, an intention recognition information vector corresponding to the above-mentioned intention recognition information, and a architectural knowledge base query information vector corresponding to the above-mentioned architectural knowledge base query information. Among them, the above-mentioned reply information optimization module includes the above-mentioned encoder layer, decoder layer, and generative adversarial network. Among them, the above-mentioned reply information vector can be a vector representing the initial reply information. The above-mentioned intention recognition information vector can be a vector representing the image recognition information. The above-mentioned architectural knowledge base query information vector can be a vector representing the architectural knowledge base query information.

[0086] The third step is to input the architectural knowledge base query information vector and the response information vector into the decoder layer to obtain an integrated response information vector. The integrated response information vector may be a vector obtained by concatenating the architectural knowledge base query information vector and the response information vector.

[0087] The fourth step is to input the above-mentioned integrated reply information vector and the above-mentioned intention recognition information vector into the above-mentioned generative adversarial network to obtain the optimized integrated reply information. Among them, the loss function value of the above-mentioned generative adversarial network can be the sum of the first value and the second value. The above-mentioned first value can be the product of the intention matching loss value and the first preset weight value. The above-mentioned second value can be the product of the sentence coherence loss value and the second preset weight value. The above-mentioned intention matching loss value can be the difference value between the intention information and the intention recognition information corresponding to the generated optimized integrated reply information. The above-mentioned difference value can be expressed as the inverse of the similarity between the intention information and the intention recognition information. The above-mentioned intention information can be the information obtained after inputting the optimized integrated reply information into a pre-trained intention classifier. The above-mentioned sentence coherence loss value can be confusion. The above-mentioned optimized integrated reply information can be the text reply information generated by the architectural knowledge question and answer model by combining the initial reply information and the architectural knowledge base query information and guided by the intention recognition information.

[0088] The fifth step is to determine the obtained optimized and integrated response information as the initial building knowledge response information.

[0089] Second, the entity concept information set included in the query information set and the initial building knowledge answer information are determined as the building knowledge answer information.

[0090] Third, the architectural knowledge reply information is sent to the client, so that the client can display the architectural knowledge reply information.

[0091] The above-mentioned embodiments of the present disclosure have the following beneficial effects: through the architectural knowledge question-and-answer system of some embodiments of the present disclosure, the real-time and accuracy of architectural knowledge reply information are improved. Specifically, the reason for the poor real-time and accuracy of architectural knowledge reply information is that: directly based on the architectural knowledge demand information, the architectural knowledge reply information is generated by a pre-trained question-and-answer model, and the question-and-answer model may be limited to the information contained in the training data. If the training data used to train the question-and-answer model contains less architectural knowledge information, the accuracy of the architectural knowledge reply information will be low. At the same time, the training of the question-and-answer model of the question-and-answer model is based on the historical data set. The architectural knowledge contained in the historical data set is static, which causes the architectural knowledge reply information input by the question-and-answer model to be unable to reflect the newly emerging architectural technology or architectural knowledge, and then, the real-time of the architectural knowledge reply information is poor. Based on this, the architectural knowledge question-and-answer system of some embodiments of the present disclosure includes: a client, a server, and a database server, wherein: first, the database server is configured to create an architectural knowledge base and an architectural knowledge graph. Then, the client is connected to the server in communication, wherein the client is configured to send the architectural knowledge demand information to the server. Thus, the architectural knowledge demand information can be sent to the server. Next, the server is configured to perform the following steps: in response to receiving the architectural knowledge demand information sent by the client, extracting the deep semantic information corresponding to the architectural knowledge demand information, wherein the deep semantic information includes: intent recognition information and a keyword set. Thus, the intent recognition information for generating initial architectural knowledge reply information can be obtained. Then, the keyword set is sent to the database server. Next, the database server is communicatively connected to the server, wherein the database server is further configured to retrieve the created architectural knowledge base and architectural knowledge graph in response to receiving the keyword set, so as to generate a query information set corresponding to the keyword set, wherein each query information in the query information set includes: an entity concept information set and architectural knowledge base query information. Thus, an entity concept information set for generating architectural knowledge reply information and architectural knowledge base query information with good real-time performance can be obtained. Next, the server is further configured to perform the following steps: Then, in response to receiving the query information set sent by the database server, the architectural knowledge requirement information, the intention recognition information and the architectural knowledge base query information included in the query information set are input into a pre-trained architectural knowledge question-answering model to obtain initial architectural knowledge answer information corresponding to the architectural knowledge requirement information. Thus, the architectural knowledge question-answering model that generates answer information for architectural knowledge questions can generate more accurate initial architectural knowledge answer information based on the architectural knowledge requirement information, the intention recognition information and the architectural knowledge base query information with better real-time performance.Afterwards, the entity concept information set included in the query information set and the initial architectural knowledge answer information are determined as architectural knowledge answer information. Thus, architectural knowledge answer information with richer answer information can be obtained. Finally, the architectural knowledge answer information is sent to the client so that the client can display the architectural knowledge answer information. Thus, the architectural knowledge answer information can be displayed through the client. Also, because the architectural knowledge base and architectural knowledge graph created by retrieving the database server are adopted, architectural knowledge base query information with good real-time performance and corresponding to the architectural knowledge demand information is generated. On the basis of the architectural knowledge demand information, the intention recognition information and the architectural knowledge base query information with good real-time performance, the architectural knowledge question-answering model that generates answer information for architectural knowledge questions generates initial architectural knowledge answer information with higher accuracy and better real-time performance. The architectural knowledge answer information is generated on the basis of the initial architectural knowledge answer information with higher accuracy and better real-time performance. The accuracy and real-time performance of the architectural knowledge answer information are improved.

[0092] Figure 2 The process 200 of some embodiments of the method for generating architectural knowledge answer information using the server included in the architectural knowledge question and answer system according to the present disclosure is shown. The method for generating architectural knowledge answer information includes the following steps:

[0093] Step 201, in response to receiving the building knowledge requirement information sent by the client, extracting deep semantic information corresponding to the building knowledge requirement information.

[0094] In some embodiments, the execution subject of generating the architectural knowledge answer information (e.g., the server included in the architectural knowledge question and answer system) may extract the deep semantic information corresponding to the architectural knowledge demand information in response to receiving the architectural knowledge demand information sent by the client. The deep semantic information includes: intention recognition information and a keyword set.

[0095] Step 202, sending the keyword set to the database server, so that the database server can retrieve the created architectural knowledge base and architectural knowledge graph to generate a query information set corresponding to the keyword set.

[0096] In some embodiments, the execution entity may send the keyword set to the database server so that the database server can retrieve the created architectural knowledge base and architectural knowledge graph to generate a query information set corresponding to the keyword set. The query information in the query information set includes: entity concept information set and architectural knowledge base query information.

[0097] Step 203, in response to receiving the query information set sent by the database server, the architectural knowledge requirement information, the intention recognition information and the architectural knowledge base query information included in the query information set are input into a pre-trained architectural knowledge question-answering model to obtain initial architectural knowledge response information corresponding to the architectural knowledge requirement information.

[0098] In some embodiments, the above-mentioned execution entity can, in response to receiving the query information set sent by the above-mentioned database server, input the above-mentioned architectural knowledge demand information, the above-mentioned intention recognition information and the architectural knowledge base query information included in the above-mentioned query information set into a pre-trained architectural knowledge question and answer model to obtain initial architectural knowledge response information corresponding to the above-mentioned architectural knowledge demand information.

[0099] Step 204: determine the entity concept information set included in the query information set and the initial building knowledge answer information as the building knowledge answer information.

[0100] In some embodiments, the execution entity may determine the entity concept information set included in the query information set and the initial building knowledge answer information as the building knowledge answer information.

[0101] Step 205: Send the architectural knowledge reply information to the client, so that the client can display the architectural knowledge reply information.

[0102] In some embodiments, the execution entity may send the architectural knowledge reply information to the client, so that the client can display the architectural knowledge reply information.

[0103] The above-mentioned embodiments of the present disclosure have the following beneficial effects: through the method for generating architectural knowledge reply information applied to the server included in the architectural knowledge question and answer system in some embodiments of the present disclosure, the real-time and accuracy of the architectural knowledge reply information are improved. Specifically, the reason for the poor real-time and accuracy of the architectural knowledge reply information is that: directly based on the architectural knowledge demand information, the architectural knowledge reply information is generated by a pre-trained question and answer model, and the question and answer model may be limited to the information contained in the training data. If the training data used to train the question and answer model contains less architectural knowledge information, the accuracy of the architectural knowledge reply information will be low. At the same time, the training of the question and answer model of the question and answer model is based on the historical data set, and the architectural knowledge contained in the historical data set is static, resulting in that the architectural knowledge reply information input by the question and answer model may not reflect the newly emerged architectural technology or architectural knowledge, and then, the real-time performance of the architectural knowledge reply information is poor. Based on this, the method for generating architectural knowledge reply information applied to the server included in the architectural knowledge question and answer system in some embodiments of the present disclosure includes: first, in response to receiving the architectural knowledge demand information sent by the client, extracting the deep semantic information corresponding to the architectural knowledge demand information, wherein the deep semantic information includes: intention recognition information and a keyword set. Thus, a keyword set for generating query information and intention recognition information for guiding the generation of reply information can be obtained. Then, the keyword set is sent to the database server so that the database server can retrieve the created architectural knowledge base and architectural knowledge graph to generate a query information set corresponding to the keyword set, wherein each query information in the query information set includes: an entity concept information set and architectural knowledge base query information. Thus, an entity concept information set for generating architectural knowledge reply information and architectural knowledge base query information with good real-time performance can be obtained. Then, in response to receiving the query information set sent by the database server, the architectural knowledge demand information, the intention recognition information and the architectural knowledge base query information included in the query information set are input into a pre-trained architectural knowledge question-and-answer model to obtain initial architectural knowledge reply information corresponding to the architectural knowledge demand information. Thus, the architectural knowledge question-and-answer model for generating reply information for architectural knowledge questions can generate initial architectural knowledge reply information with higher accuracy based on the architectural knowledge demand information, the intention recognition information and the architectural knowledge base query information with good real-time performance. The entity concept information set included in the query information set and the initial architectural knowledge reply information are determined as architectural knowledge reply information. Thus, architectural knowledge reply information with richer reply information can be obtained. Finally, the architectural knowledge reply information is sent to the client so that the client can display the architectural knowledge reply information.The architectural knowledge question-answering model that generates answer information for architectural knowledge questions based on architectural knowledge demand information, the above-mentioned intention recognition information, and architectural knowledge base query information with good real-time performance generates initial architectural knowledge answer information with higher accuracy and better real-time performance. The architectural knowledge answer information is generated based on the initial architectural knowledge answer information with higher accuracy and better real-time performance. The accuracy and real-time performance of the architectural knowledge answer information are improved.

[0104] The above descriptions are only some preferred embodiments of the present disclosure and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present disclosure is not limited to the technical solutions formed by a specific combination of technical features, but should also cover other technical solutions formed by any combination of technical features or their equivalent features without departing from the inventive concept. For example, a technical solution formed by replacing a feature with a technical feature with similar functions disclosed in the embodiments of the present disclosure (but not limited to).

Claims

1. An architectural knowledge question-answering system, comprising: Client, server, database server, where: The database server is configured to create a building knowledge base and a building knowledge graph; The client is communicatively connected to the server, wherein the client is configured to send the building knowledge requirement information to the server; The server is configured to perform the following steps: In response to receiving the building knowledge requirement information sent by the client, extracting deep semantic information corresponding to the building knowledge requirement information, wherein the deep semantic information includes: intent recognition information and a keyword set; sending the keyword set to the database server; The database server is communicatively connected to the server, wherein the database server is further configured to retrieve the created architectural knowledge base and architectural knowledge graph in response to receiving the keyword set to generate a query information set corresponding to the keyword set, wherein each query information in the query information set includes: an entity concept information set, architectural knowledge base query information; The server is further configured to perform the following steps: In response to receiving the query information set sent by the database server, the architectural knowledge requirement information, the intention recognition information and the architectural knowledge base query information included in the query information set are input into a pre-trained architectural knowledge question-answering model to obtain initial architectural knowledge reply information corresponding to the architectural knowledge requirement information, wherein the architectural knowledge question-answering model comprises: a question-answering model and a reply information optimization module; and the server is further configured to input the architectural knowledge requirement information, the intention recognition information and the architectural knowledge base query information included in the query information set into the pre-trained architectural knowledge question-answering model through the following steps to obtain initial architectural knowledge reply information corresponding to the architectural knowledge requirement information, including: Inputting the building knowledge demand information into the question-answering model to obtain initial answer information corresponding to the building knowledge demand information; Inputting the initial reply information, the intention recognition information and the building knowledge base query information into the encoder layer of the reply information optimization module, obtaining a reply information vector corresponding to the initial reply information, an intention recognition information vector corresponding to the intention recognition information, and a building knowledge base query information vector corresponding to the building knowledge base query information, wherein the reply information optimization module includes the encoder layer, the decoder layer, and the generative adversarial network; Inputting the architectural knowledge base query information vector and the response information vector into the decoder layer to obtain an integrated response information vector; Inputting the integrated response information vector and the intention recognition information vector into the generative adversarial network to obtain optimized integrated response information; Determine the obtained optimized integrated answer information as initial building knowledge answer information; Determining the entity concept information set included in the query information set and the initial building knowledge answer information as building knowledge answer information; The architectural knowledge reply information is sent to the client so that the client displays the architectural knowledge reply information.

2. The architectural knowledge question-answering system according to claim 1, wherein: The database server is further configured to create a building knowledge base and a building knowledge graph through the following steps: Acquire a building knowledge information set from a preset network storage device; Create an initial database based on preset demand information; Storing the building knowledge information set in the initial database to update the initial database; identifying the updated initial database as a building knowledge base; Performing entity recognition on the architectural knowledge information set to obtain an entity information set; Determine each entity information in the entity information set as each graph node; Based on the entity information, the building knowledge information set is subjected to entity relationship extraction processing to obtain entity relationship information, wherein each entity relationship information in the entity relationship information includes: starting point entity information, relationship information, and pointing entity information; For each entity relationship information in the entity relationship information, the following steps are performed: Determine each graph node in the graph nodes that corresponds to the starting entity information and the pointing entity information included in the entity relationship information as a graph node group; Determine the relationship information included in the entity relationship information as edge information between graph nodes in the graph node group; Generate an initial architectural knowledge graph based on each graph node group with determined edge information; Based on the entity information and the initial architectural knowledge graph, an architectural knowledge graph is generated.

3. The architectural knowledge question-answering system according to claim 2, wherein: The database server is further configured to generate a building knowledge graph based on the entity information and the initial building knowledge graph through the following steps, including: For each piece of entity information, perform the following steps: Acquire entity concept information corresponding to the entity information from the preset network storage device; Determine a graph node corresponding to the entity information in the initial architectural knowledge graph as a target graph node; Adding the entity concept information as graph node attribute information to the target graph node to update the initial building knowledge graph; The updated initial architectural knowledge graph is determined as the architectural knowledge graph.

4. The architectural knowledge question-answering system according to claim 1, wherein: The server is further configured to extract deep semantic information corresponding to the building knowledge requirement information in response to receiving the building knowledge requirement information sent by the client through the following steps, including: Inputting the building knowledge demand information into a pre-trained keyword extraction model to obtain a keyword set corresponding to the building knowledge demand information; Inputting the building knowledge requirement information into a pre-trained intention recognition model to obtain intention recognition information corresponding to the building knowledge requirement information; The keyword set and the intent recognition information are determined as deep semantic information.

5. The architectural knowledge question-answering system according to claim 4, wherein: The server is further configured to input the building knowledge requirement information into a pre-trained intention recognition model through the following steps to obtain intention recognition information corresponding to the building knowledge requirement information, including: Inputting the building knowledge demand information into the input layer of the intention recognition model to obtain a word segmentation sequence corresponding to the building knowledge demand information, wherein the intention recognition model includes the input layer, the word embedding layer, the relative position encoding layer, the stacked attention module, and the intention classification layer, the stacked attention module includes various intention extraction modules, and each of the various intention extraction modules includes a self-attention layer and a feedforward neural network layer; Inputting the word segmentation sequence into the word embedding layer to obtain a word vector sequence corresponding to the word segmentation sequence; Inputting the word vector sequence into the relative position encoding layer to obtain a relative position embedding vector sequence; Inputting the relative position embedding vector sequence into the stacked attention module to obtain a weighted transformation word vector sequence corresponding to the relative position embedding vector sequence; The weighted transformation word vector sequence is input into the intention classification layer to obtain the intention recognition information corresponding to the weighted transformation word vector sequence.

6. A method for generating architectural knowledge answer information, applied to a server included in the architectural knowledge question-answering system according to any one of claims 1 to 5, the method comprising: In response to receiving the building knowledge requirement information sent by the client, extracting deep semantic information corresponding to the building knowledge requirement information, wherein the deep semantic information includes: intent recognition information and a keyword set; The keyword set is sent to the database server, so that the database server can retrieve the created architectural knowledge base and architectural knowledge graph to generate a query information set corresponding to the keyword set, wherein each query information in the query information set includes: an entity concept information set and architectural knowledge base query information; In response to receiving the query information set sent by the database server, inputting the architectural knowledge requirement information, the intention recognition information and the architectural knowledge base query information included in the query information set into a pre-trained architectural knowledge question-answering model to obtain initial architectural knowledge answer information corresponding to the architectural knowledge requirement information; Determining the entity concept information set included in the query information set and the initial building knowledge answer information as building knowledge answer information; The architectural knowledge reply information is sent to the client so that the client displays the architectural knowledge reply information.

Images