Investment estimation data generation method and equipment for project cost consultation, and medium

Abstract

The invention relates to the technical field of project cost consultation, and provides an investment estimation data generation method and device for project cost consultation and a medium. The investment estimation data generation method comprises the steps of embedding building category features corresponding to a target project into target cost consultation data; the target cost consultation data is input into a pre-trained to-be-evaluated data extraction model, to-be-evaluated data corresponding to the target project is obtained, the type of the to-be-evaluated data corresponds to the project type of the target project, and the to-be-evaluated data extraction model comprises a multi-head attention network layer; determining an index value of a preset index item corresponding to the target project based on the to-be-evaluated data; and generating investment estimation data corresponding to the target project based on the index values of the preset index items. By adopting the technical scheme, the to-be-evaluated data corresponding to the target project can be extracted by using the to-be-evaluated data extraction model, so that the dependence of an investment estimation process on manpower can be reduced, and the accuracy of an investment estimation result can be ensured.

Description

Technical Field

[0001] The present application relates to the technical field of engineering cost consulting, and in particular to a method, device and medium for generating investment estimation data for engineering cost consulting. Background Art

[0002] Engineering cost consulting is a key link in engineering projects, among which investment estimation is an important step in determining project budget and investment return. In related technologies, there are many methods for investment estimation, such as coefficient estimation method, budget estimation method, etc.

[0003] Taking the rough estimate method as an example, in the relevant technology, it is necessary to manually and tabulate the various indicator items and their corresponding indicator values ​​in the cost consulting data of the target project, and then make an investment estimate for the target project based on the indicator values ​​corresponding to each indicator item.

[0004] However, the accuracy of investment estimation results depends on the accuracy of statistical indicators and indicator values, and the accuracy of indicators and indicator values ​​manually and in tables depends on manual experience, which leads to the problem of unstable accuracy of investment estimation results in related technologies. Summary of the invention

[0005] In order to help ensure the accuracy of investment estimation results, the present application provides a method, device and medium for generating investment estimation data for engineering cost consulting.

[0006] In a first aspect, the present application provides a method for generating investment estimation data for engineering cost consulting, which adopts the following technical solution: A method for generating investment estimation data for engineering cost consulting, the method comprising: Obtain the target cost consulting data corresponding to the target project; Embedding a building category feature corresponding to the target project into the target cost consulting data, wherein the building category feature is used to indicate the project category of the target project; Inputting the target cost consulting data into a pre-trained data extraction model to be evaluated, obtaining data to be evaluated corresponding to the target project, wherein the type of the data to be evaluated corresponds to the project category of the target project, and the data extraction model to be evaluated includes a multi-head attention network layer; Determine an index value of a preset index item corresponding to the target project based on the data to be evaluated, wherein the preset index item is related to the project category of the target project; The investment estimation data corresponding to the target project is generated based on the indicator value of the preset indicator item.

[0007] By adopting the above technical solution, the data to be evaluated corresponding to the target project can be extracted using the data extraction model, and the indicator values ​​corresponding to the preset indicator items can be further extracted based on the data to be evaluated. In this way, the automatic extraction of indicator items and indicator values ​​in the cost consulting data and the automatic generation of investment estimation data can be realized, thereby reducing the dependence of the investment estimation process on manual labor, which can help ensure the accuracy of the investment estimation results.

[0008] Optionally, the data extraction model to be evaluated is trained based on the following method: Obtain historical engineering cost consulting data for different preset engineering categories; For each historical engineering cost consulting data, embedding the building category feature corresponding to the historical engineering cost consulting data into the historical engineering cost consulting data, and determining the sample data to be evaluated in the historical cost evaluation data based on the type of data to be evaluated corresponding to the engineering category of the historical engineering cost consulting data; Determine each of the historical engineering cost consulting data and the engineering category and sample data to be evaluated corresponding to the historical engineering cost consulting data as a set of sample data; A training data set is constructed based on the sample data to train the pre-constructed initial network model in batches to obtain the data extraction model to be evaluated, and each of the training data sets includes the sample data of at least two engineering categories.

[0009] In the above technical solution, by configuring at least two categories of sample data in the training set, the trained data extraction model for to-be-evaluated can extract the data to-be-evaluated from the cost consulting data in combination with the actual conditions of different engineering categories, thereby helping to improve the accuracy of the extracted data to be evaluated.

[0010] Optionally, the number of attention heads of the multi-head attention network layer matches the number of data types to be evaluated corresponding to each preset engineering category.

[0011] By adopting the above technical solution, it can be ensured that each type of data to be evaluated in the target cost consulting data can be effectively extracted, which can help improve the accuracy of cost evaluation.

[0012] Optionally, the attention network layer includes a first attention head and a second attention head, the first attention head is used to extract evaluation data types common to at least two of the preset engineering categories; the second attention head is used to extract evaluation data types used by a specific preset engineering category.

[0013] By adopting the above technical solution, it can help to reduce the number of attention heads in the multi-head attention network while ensuring that each type of data to be evaluated in the target cost consulting data can be effectively extracted, which can help reduce the size of the data extraction model to be evaluated.

[0014] Optionally, the multi-head attention network layer also includes a category prediction module; The category prediction module is used to predict the engineering category to which the input data belongs, so as to guide the multi-head attention network to adjust the calculation strategy of the attention head based on the working category.

[0015] By adopting the above technical solution, the corresponding engineering category can be identified before extracting the data to be evaluated, so that the parameters of the multi-head attention network layer can be adjusted in combination with the engineering category, which can help to extract the data to be evaluated of different engineering categories in a targeted manner, and further help to improve the accuracy of the extracted data to be evaluated.

[0016] Optionally, before determining the index values ​​of each preset index item of the engineering category of the target engineering based on the data to be evaluated, the method further includes: For each of the data to be evaluated, determining whether the data to be evaluated meets preset format requirements; In the case where the data to be evaluated does not meet the format requirements, searching the target cost consulting data for associated data of the data to be evaluated based on the data to be evaluated; The data to be evaluated is modified based on the associated data.

[0017] By adopting the above technical solution, it is possible to detect whether the evaluation data is accurate, and when it is detected that there is missing information in the data to be evaluated, the data to be evaluated can be corrected in combination with the content related to the data to be evaluated in the target cost consulting data, which can help reduce the problem of erroneous indicator extraction caused by missing information in the data to be evaluated, and further help improve the accuracy of the investment estimation results.

[0018] Optionally, before generating the investment estimation data corresponding to the target project based on the indicator value of the preset indicator item, the method further includes: For each of the indicator values, determining whether the indicator value belongs to a corresponding data range, the data range being pre-set based on a type of an indicator item corresponding to the indicator value; If the indicator value does not fall within the data range, determining whether to modify the indicator value; In the case where it is determined to correct the indicator value, the indicator value is corrected based on a correction method corresponding to the indicator value, and the correction method is pre-set based on a type of an indicator item corresponding to the indicator value.

[0019] By adopting the above technical solution, the indicator value can be checked and processed for abnormality before investment estimation, which can help reduce the impact of abnormal indicator values ​​on investment estimation results, and further help improve the accuracy of investment estimation results.

[0020] Optionally, after generating the investment estimation data corresponding to the target project, the method further includes: Determining the level interval to which the investment estimation data belongs; When the level interval indicates that the investment estimation data exceeds the budget, determining the indicator item exceeding the budget from each of the indicator items; Determining abnormal indicator items from the indicator items exceeding the budget; Output the abnormal indicator item and the indicator value corresponding to the abnormal indicator item.

[0021] By adopting the above technical solution, after determining the investment estimation data, it is possible to determine the level interval to which the investment estimation data belongs, and when the level interval indicates that the investment estimation data exceeds the budget, the abnormal indicator items and corresponding indicator values ​​are determined and output in combination with the actual conditions of each indicator, thereby guiding the staff to handle the abnormal indicator items accordingly when the investment estimation is abnormal.

[0022] In a second aspect, the present application provides an electronic device, which adopts the following technical solution: An electronic device, comprising: at least one processor; Memory; At least one application, wherein the at least one application is stored in a memory and configured to be executed by at least one processor, and the at least one application is configured to: execute any one of the investment estimation data generation methods for engineering cost consulting provided in the first aspect.

[0023] In a third aspect, the present application provides a computer-readable storage medium, which adopts the following technical solution: A computer-readable storage medium stores a computer program, which, when executed in a computer, causes the computer to execute any one of the investment estimation data generation methods for engineering cost consulting provided in the first aspect.

[0024] In summary, the present application includes at least one of the following beneficial technical effects: 1. By using the data extraction model to extract the data to be evaluated corresponding to the target project, and further extracting the indicator values ​​corresponding to the preset indicator items based on the data to be evaluated, the automatic extraction of indicator items and indicator values ​​in the cost consulting data and the automatic generation of investment estimation data can be realized, thereby reducing the dependence of the investment estimation process on manual labor, and thus helping to ensure the accuracy of the investment estimation results.

[0025] 2. Since the building category characteristics corresponding to the engineering category of the target project are embedded in the target cost consulting data, and the type of data to be evaluated extracted by the data extraction model corresponds to the category of the target project, and the preset indicator items are also related to the engineering category of the target project, the engineering category of the target project can be fully considered in the process of indicator extraction, which can help improve the accuracy of the extracted indicators. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a flow chart of a method for generating investment estimation data for engineering cost consulting provided in an embodiment of the present application; Figure 2 It is a flowchart of a method for training a model for extracting data to be evaluated provided in an embodiment of the present application; Figure 3 It is a flowchart of a method for correcting data to be evaluated provided in an embodiment of the present application; Figure 4 It is a flow chart of an indicator value correction method provided in an embodiment of the present application; Figure 5 This is a flow chart of a method for determining abnormal indicator values ​​provided in an embodiment of the present application; Figure 6 It is a flow chart of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solutions and advantages of this application more clear, the following Figure 1-6 It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0028] The present application embodiment discloses a method for generating investment estimation data for engineering cost consulting. Figure 1 , investment estimate data generation includes: Step 101, obtaining target cost consulting data corresponding to a target project.

[0029] Among them, cost consulting data refers to data that can be used to estimate project investment. Specifically, project cost consulting data can include at least one text material related to the project, such as project description, bill of quantities, material usage report, design instructions, etc. These materials usually contain a large amount of text, and most of them are formatted text and introductory text. The information that is valuable for investment estimation is scattered and difficult to use directly.

[0030] Furthermore, after obtaining the text data related to the project, the text data is preprocessed to obtain the target cost consulting data. Specifically, the method of preprocessing the text data includes but is not limited to deleting the contents of the chapters in the text data that are not related to the project investment estimation, performing text recognition on the pictures in the text data and deleting the pictures after retaining the recognition results, and extracting the data in the table in the text data and deleting the table after retaining the extraction results. This can help to extract information from the target cost consulting data later.

[0031] Step 102: embed the building category features corresponding to the target project into the target cost consulting data.

[0032] The building category feature is used to indicate the project category of the target project. Specifically, the project category may include residential building category, commercial building category, industrial building category, public building category, etc. Considering that the investment influencing factors corresponding to different project categories may be different, the project category corresponding to the target project needs to be considered in the process of investment estimation.

[0033] In one example, the building category feature is additionally added according to the engineering category of the target project. Specifically, different building categories may correspond to different category identifiers, and in this case, the corresponding category identifier can be added to the target cost consulting data according to the building category corresponding to the target project.

[0034] In actual implementation, the building category features are automatically extracted from the target cost consulting data, for example, they are automatically extracted from the project name and project introduction. At this time, the extracted features can be directly embedded in the preset position of the target cost consulting data.

[0035] Step 103: input the target cost consulting data into the pre-trained to-be-evaluated data extraction model to obtain the to-be-evaluated data corresponding to the target project.

[0036] The type of data to be evaluated corresponds to the engineering category of the target project, and the data extraction model to be evaluated includes a multi-head attention network layer. Specifically, the data to be evaluated can be actual values, such as building area, number of floors, or construction requirements, such as safety requirements, construction process requirements, etc. As long as it will have an impact on the project investment, this embodiment does not limit the type of data to be evaluated.

[0037] In this embodiment, the types of data to be evaluated corresponding to different engineering categories are pre-set, and the types of data to be evaluated corresponding to different engineering categories are different. Specifically, taking the engineering categories including residential buildings, commercial buildings, and industrial buildings as an example, the difference in the types of data to be evaluated corresponding to the three is that the types of data to be evaluated corresponding to the residential building category can include residential scale data, residential material data, etc.; the types of data to be evaluated corresponding to the commercial building category can include commercial customization data, commercial material data, commercial supporting facilities data, etc.; the data to be evaluated corresponding to the industrial building category can include safety system data, green material data, etc.

[0038] In this embodiment, the evaluation data extraction model can extract different types of data to be evaluated in the target cost consulting data through a multi-head attention network layer, so that valuable information for investment estimation can be extracted from massive text data, which can help to make a full investment estimate for the target project.

[0039] In some embodiments, the number of attention heads in the multi-head attention network layer matches the number of evaluation data types corresponding to each preset engineering category. In one example, the attention heads correspond one-to-one to the types of data to be evaluated, so that each type of data to be evaluated in the target cost consulting data can be effectively extracted, which can help improve the accuracy of cost evaluation. In actual implementation, the same attention head can also be used to extract the data to be evaluated corresponding to multiple similar data types to be evaluated, which can help improve the data extraction efficiency of the data extraction model to be evaluated.

[0040] Furthermore, since different engineering categories may have the same type of data to be evaluated, such as the number of floors, building height, building area, etc., although their expressions in different cost consulting data may be different, the essential content is the same. At this time, we can correspond the same type of data to be evaluated in different engineering categories to the same attention head. At this time, the attention network layer can include the first attention head and the second attention head.

[0041] Among them, the first attention head is used to extract the evaluation data type common to at least two preset engineering categories; the second attention head is used to extract the evaluation data type used by a specific preset engineering category.

[0042] By adopting the above technical solution, it can help to reduce the number of attention heads in the multi-head attention network while ensuring that each type of data to be evaluated in the target cost consulting data can be effectively extracted, which can help reduce the size of the data extraction model to be evaluated.

[0043] In some embodiments, the multi-head attention network layer also includes a category prediction module, which is used to predict the engineering category to which the input data belongs, so as to guide the multi-head attention network to adjust the calculation strategy of the attention head based on the engineering category. For example, it is determined to enable the attention head corresponding to the data type to be evaluated corresponding to the engineering category.

[0044] Among them, the category prediction module can identify the building category characteristics from the target cost consulting data and determine the project category based on the building category characteristics.

[0045] In one example, the building category includes an additional category identifier, and the project category can be determined directly based on the category identifier.

[0046] In another example, the building category feature includes data extracted from the target cost consulting data, and the data can also be analyzed to determine the project category. In one example, the category prediction module is a forward network layer set before the multi-head attention network layer for identifying the building category, which can be implemented by a pre-trained classification network. The input of the network is the building category feature. The forward network layer can adjust the parameters of the multi-head attention network according to the input prediction to adjust the calculation strategy of the attention head of the multi-head attention network.

[0047] Step 104: determine the index value of the preset index item corresponding to the target project based on the data to be evaluated.

[0048] Among them, the preset indicator items are related to the engineering category of the target project. In actual implementation, the preset indicators corresponding to different engineering categories may be different. In one example, the correspondence between the data to be evaluated and the preset indicator items is preset. For example, if the data to be evaluated is residential scale data, the corresponding indicator items may be floor area, total area, number of floors, etc.

[0049] In one example, determining the index value of the preset index item corresponding to the target project based on the data to be evaluated includes: parsing the data to be evaluated to obtain the index item to be processed and its corresponding index value; matching the index item to be processed with the preset index item to obtain the index value of the index item to be processed.

[0050] Among them, the method of parsing the data to be evaluated can use a semantic recognition model to parse the data to be evaluated, and obtain the indicators to be processed and their corresponding indicator values. For example: the data to be evaluated includes a building area of ​​1,000 square meters. At this time, the indicator item to be processed is the building area, and its corresponding indicator value is 1,000 square meters. For another example: the data to be evaluated is that the project is a three-story residential building with a total construction area of ​​3,000 square meters, using ordinary household building materials. At this time, the indicators to be processed are the number of floors, the total construction area, and the type of building materials, and the corresponding indicator values ​​are three floors, 3,000 square meters, and ordinary household.

[0051] The method of matching the indicator item to be processed with the preset indicator item can be to calculate the textual and / or semantic similarity between each preset indicator item and the indicator item to be processed, determine the preset indicator item with the highest similarity as the preset indicator item corresponding to the indicator item to be processed, and determine the indicator value of the indicator item to be processed as the indicator value of the preset indicator item corresponding to it.

[0052] In one example, each preset index item may be sorted in descending order of textual similarity with the index item to be processed, and the semantic similarity between the first-ranked preset index item and the second-ranked preset index item may be determined.

[0053] When the semantic similarity between the first-ranked preset index item and the second-ranked preset index item is less than a preset semantic similarity threshold, the first-ranked preset index item is determined as the semantic index item corresponding to the index item to be processed.

[0054] When the semantic similarity between the first-ranked preset indicator item and the second-ranked preset indicator item is greater than or equal to the preset semantic similarity threshold, for each preset indicator item, the weighted average of the textual and semantic similarities between the preset indicator item and the indicator item to be processed is determined as the similarity between the preset indicator item and the indicator item to be processed, and the preset indicator item with the highest similarity to the indicator item to be processed is determined as the preset indicator item corresponding to the indicator item to be processed.

[0055] In actual implementation, the preset index item and the data to be evaluated can also be directly matched for similarity, so as to determine the preset index item corresponding to the data to be evaluated, and then determine the index value corresponding to the preset index item. In the case where the data to be evaluated may include multiple preset index items, the position of the matched preset index item in the data to be evaluated can be further determined during the matching process, which can help to accurately extract the index values ​​corresponding to different index items from the data to be evaluated.

[0056] Step 105: Generate investment estimation data corresponding to the target project based on the indicator values ​​of the preset indicator items.

[0057] In one example, generating investment estimation data corresponding to a target project based on the indicator values ​​of preset indicator items includes: comprehensively determining the investment estimation data based on the indicator values ​​of preset indicator items and the weights corresponding to the preset indicator items. In actual implementation, the weights corresponding to the preset indicator items can be determined based on the importance of the indicator items to the project. For example, the weight of the safety indicator is higher than the weight of the indicator corresponding to the supporting facilities. In this way, in the process of investment estimation, more focus can be placed on the key tasks of the project, which can help improve the accuracy of the investment estimation data.

[0058] In one example, the investment estimation data is determined based on the index value of the preset index item and the weight corresponding to the preset index item, including: for each index item, the estimated cost corresponding to the index item is determined by the index value corresponding to the index item and the unit cost corresponding to the index item; based on the weight corresponding to the preset index item, the weighted sum of the estimated costs of each preset index item is determined as the investment estimation data. Among them, the unit cost corresponding to the index item is predetermined. In actual implementation, the unit cost can be extracted from the historical engineering final settlement data. Furthermore, the unit cost needs to be combined with the actual situation of the project construction time and location, such as: considering the differences in different time periods and different regions, which can help improve the accuracy of the unit cost, and then help improve the accuracy of the investment estimation data finally generated.

[0059] In actual implementation, the investment estimation data may also be determined directly by combining the sum of the estimated costs corresponding to the various indicator items without considering the weights of the various indicator items.

[0060] Furthermore, in the process of investment estimation, other indicators such as management cost indicators can be further determined based on preset indicator items, and the investment can be further estimated in combination with other indicators, such as adjusting the investment estimation data calculated in the above manner in combination with the management cost indicator.

[0061] The implementation principle of the investment estimation data generation method for engineering cost consulting in the embodiment of the present application is as follows: obtaining the target cost consulting data corresponding to the target project; embedding the building category feature corresponding to the target project into the target cost consulting data, and the building category feature is used to indicate the engineering category of the target project; inputting the target cost consulting data into the pre-trained data extraction model to be evaluated, obtaining the data to be evaluated corresponding to the target project, the type of the data to be evaluated corresponds to the engineering category of the target project, and the data extraction model to be evaluated includes a multi-head attention network layer; determining the index value of the preset index item corresponding to the target project based on the data to be evaluated, and the preset index item is related to the engineering category of the target project; generating the investment estimation data corresponding to the target project based on the index value of the preset index item. By adopting the above technical solution, the data extraction model to be evaluated can be used to extract the data to be evaluated corresponding to the target project, and the index value corresponding to the preset index item can be further extracted based on the data to be evaluated, so that the automatic extraction of the index items and the index values ​​in the cost consulting data and the automatic generation of the investment estimation data can be realized, thereby reducing the dependence of the investment estimation process on manual labor, and thus helping to ensure the accuracy of the investment estimation results.

[0062] At the same time, since the data to be evaluated is extracted from the cost consulting data, and then the indicator value corresponding to the preset indicator item is determined based on the data to be evaluated, this can help reduce the impact of a large amount of text data irrelevant to investment in the cost consulting data on indicator extraction, and thus can help improve the efficiency and accuracy of indicator extraction.

[0063] At the same time, since the data extraction model for evaluation includes a multi-head attention network layer, pre-training can be used to enable different attention heads of the multi-head attention network layer to focus on different types of data to be evaluated, which can help improve the accuracy of data extraction for evaluation.

[0064] At the same time, since the building category characteristics corresponding to the engineering category of the target project are embedded in the target cost consulting data, and the type of data to be evaluated extracted by the data extraction model corresponds to the category of the target project, and the preset indicator items are also related to the engineering category of the target project, the engineering category of the target project can be fully considered in the process of indicator extraction, which can help improve the accuracy of the extracted indicators.

[0065] In some embodiments, reference Figure 2 , the data extraction model to be evaluated is trained based on the following method: Step 201, obtaining historical engineering cost consulting data of different preset engineering categories.

[0066] The historical engineering cost consulting data is cost consulting data corresponding to the engineering projects carried out in the past. The content of the historical engineering cost consulting data is the same as the content of the target cost consulting data in the above step 101, and will not be repeated here.

[0067] The preset engineering categories are set according to actual needs. For example, if the data extraction model to be evaluated needs to be able to extract engineering cost consulting data for residential building categories, commercial building categories, industrial building categories, and public building categories, then it is necessary to obtain the historical cost engineering consulting data corresponding to the residential building categories, commercial building categories, industrial building categories, and public building categories respectively.

[0068] Step 202: for each historical engineering cost consulting data, embed the building category feature corresponding to the historical engineering cost consulting data into the historical engineering cost consulting data, and determine the sample data to be evaluated in the historical cost evaluation data based on the type of data to be evaluated corresponding to the engineering category of the historical engineering cost consulting data.

[0069] Specifically, the method of embedding the building category features corresponding to the historical engineering cost consulting data into the historical engineering cost consulting data is as described in the above step 102, which will not be described in detail here.

[0070] The types of data to be evaluated corresponding to different engineering categories are pre-set by experienced evaluators based on the evaluation, and the sample data to be evaluated in the historical cost evaluation data are obtained by manual labeling.

[0071] Step 203: determine each historical engineering cost consulting data and the engineering category and sample data to be evaluated corresponding to the historical engineering cost consulting data as a group of sample data.

[0072] Step 204 , construct a training data set based on the sample data to train the pre-constructed initial network model in batches to obtain a data extraction model to be evaluated.

[0073] Each training data set includes sample data of at least two engineering categories. In actual implementation, the historical engineering data can be divided into several data sets, and the initial network model can be trained using these data sets respectively until a preset number of training times is reached or the initial network model converges.

[0074] In one example, each training data set includes sample data of all engineering categories, which can help balance the training effects of different engineering categories. Furthermore, the same sample data can belong to different training data sets, which can help solve the problem of unbalanced number of sample data of different engineering categories. Furthermore, the overlap of sample data between different training data sets is no more than 50%, which can help enrich the combination of sample data and thus help improve the training effect of the model.

[0075] Optionally, the pre-trained initial network model is trained in batches based on the sample training data set to obtain the data extraction model to be evaluated, including: for each group of sample data in the sample training data set, the historical engineering cost consulting data in the group of sample data is input into the initial network model, and the output of the initial model is evaluated based on the engineering category corresponding to the historical engineering cost consulting data and the data to be evaluated to obtain the evaluation prediction loss and category prediction loss of the initial neural network model under the group of sample data; the average value of the evaluation prediction loss corresponding to the sample data with the same preset engineering category in the sample data set is determined as the evaluation prediction loss of the initial network model under the preset engineering category; the average value of the category prediction loss corresponding to each sample data is determined as the category prediction loss of the initial network model; the weighted average value of the evaluation prediction loss and the category prediction loss of the initial network model on each preset engineering category is determined as the loss value corresponding to the initial network model; when the loss value corresponding to the initial network model is greater than the preset loss value threshold, the model parameters of the initial network model are adjusted, and the initial network model is continued to be trained using the sample training data set; when the loss value corresponding to the initial network model is less than or equal to the preset loss value threshold, it is determined that the training of the initial network model is completed to obtain the data extraction model to be evaluated.

[0076] In one example, the loss function used in the training process of the initial network model is expressed as follows: L0=w1×L1+……+w n ×L n +w n ×L c Among them, L0 is the loss value corresponding to the initial network model; L1 is the evaluation prediction loss of the initial network model under the first preset engineering category, and L n is the evaluation prediction loss of the initial network model under the nth preset engineering category, where n is the number of preset engineering categories; w1 is the weight corresponding to the first preset engineering category, w n is the weight corresponding to the nth preset engineering category; L c is the category prediction loss of the initial network model, L c The weight corresponding to the category classification.

[0077] In actual implementation, the weights of different preset engineering categories can be set according to actual conditions. For example, the more types of data to be evaluated corresponding to the preset engineering categories, the greater the weight.

[0078] Furthermore, the weights corresponding to different preset engineering categories can be dynamically adjusted during the training process. For example, the size of the weights can be adjusted based on the relative size of the evaluation prediction loss. Specifically, the weights corresponding to the preset engineering categories with relatively large evaluation prediction losses are increased, and the weights corresponding to the preset engineering categories with relatively small evaluation prediction losses are reduced. This can help increase the attention paid to engineering categories with larger evaluation prediction losses during model training, thereby enabling the model to balance the learning of different engineering categories.

[0079] In the above implementation, sample data can be generated based on historical engineering cost consulting data, and a training data set can be constructed based on the sample data to train the initial network model, and each training data set includes sample data of at least two engineering categories. In this way, the trained data extraction model for to-be-evaluated data can extract the data to be evaluated from the cost consulting data in combination with the actual conditions of different engineering categories, which can help improve the accuracy of the extracted data to be evaluated.

[0080] In practice, it is found that information may be missing when extracting data using the model to be evaluated. At this time, directly extracting indicators from the data to be evaluated may lead to problems such as failure to match corresponding indicator items, wrong indicator items being matched, and indicator values ​​being unavailable, which will affect the accuracy of investment estimates.

[0081] In order to help solve the above problems, the following implementation methods are further provided: refer to Figure 3 , step 104, before determining the index value of the preset index item corresponding to the target project based on the data to be evaluated, the following steps are also included: Step 301: for each piece of data to be evaluated, determine whether the data to be evaluated meets preset format requirements.

[0082] Optionally, the format requirements include that the data to be evaluated includes indicator items and indicator values. Accordingly, determining whether the data to be evaluated meets the corresponding format requirements includes: parsing the data to be evaluated to determine whether the indicator items and indicator values ​​exist in the data to be evaluated; if the indicator items and / or indicator values ​​do not exist in the data to be evaluated, determining that the data to be evaluated does not meet the corresponding format requirements.

[0083] Furthermore, when there are indicator items and / or indicator values ​​in the data to be evaluated, it is determined whether the indicator items and the indicator values ​​match; if the indicator items and the indicator values ​​do not match, it is determined that the data to be evaluated does not meet the corresponding format requirements.

[0084] The matching between indicator items and indicator values ​​means that each indicator item has a corresponding indicator value, and each indicator value has a corresponding indicator item. Specifically, the matching relationship between indicator values ​​and indicator items can be obtained based on semantic analysis. For example, if the data to be evaluated is "the project is a three-story residential building, 3,000 square meters, using ordinary household building materials", the number of floors corresponds to three floors, and the type of building materials corresponds to ordinary household buildings, but there is no corresponding indicator item for 3,000 square meters. Therefore, the data to be evaluated does not meet the corresponding format requirements.

[0085] Furthermore, when the indicator items and the indicator values ​​match, it is determined whether the unit of each indicator value is clear; when there is an indicator value with unclear unit, it is determined that the data to be evaluated does not meet the corresponding format requirements. For example, the data to be evaluated includes building area 1000, and the unit of the indicator value 1000 is unclear. Therefore, the data to be evaluated does not meet the corresponding format requirements.

[0086] Step 302: When the data to be evaluated does not meet the format requirements, the associated data of the data to be evaluated is retrieved from the target cost consulting data based on the data to be evaluated.

[0087] When the data to be evaluated meets the format requirements, step 104 is executed.

[0088] The associated data refers to data related to the data to be evaluated. Specifically, since the data to be evaluated is extracted from the target cost consulting data, the data corresponding to the data to be evaluated can be retrieved from the target cost data.

[0089] Optionally, retrieving the associated data of the data to be evaluated in the target cost consulting data based on the data to be evaluated includes: determining paragraphs in the cost consulting data whose similarity with the data to be evaluated is higher than a preset similarity threshold as the associated data of the data to be evaluated.

[0090] Furthermore, when there is no data in the cost consulting data whose similarity with the data to be evaluated is higher than a preset similarity threshold, the paragraphs in the cost consulting data containing the index items and index values ​​in the data to be evaluated are determined as associated data of the data to be evaluated.

[0091] Furthermore, after determining the paragraph corresponding to the data to be evaluated, the context of the paragraph, such as a preset number of paragraphs before and after the paragraph, can be determined as associated data of the data to be evaluated, which can help improve the comprehensiveness of the associated data.

[0092] Step 303: modify the data to be evaluated based on the associated data.

[0093] Optionally, the data to be evaluated is modified based on the associated data, including: searching for supplementary data in the associated data based on the reason why the data to be evaluated does not meet the format requirements; and supplementing the data to be evaluated based on the supplementary data so that the data to be evaluated meets the format requirements.

[0094] The reason why the data to be evaluated does not meet the format requirements is determined based on step 301. For example, when the reason for not meeting the format requirements is that the indicator item does not exist, the indicator item corresponding to the indicator value is searched in the associated data based on the indicator value in the data to be evaluated, and the found indicator item is supplemented to the data to be evaluated as supplementary data.

[0095] For another example, when the reason for not meeting the format requirements is that the unit of the indicator value is unclear, the unit corresponding to the indicator value is searched in the associated data based on the indicator value with unclear unit, and the found unit is added to the data to be evaluated as supplementary data.

[0096] Further, after the data to be evaluated is modified based on the associated data, the process returns to step 301 , so that when the modified data to be evaluated still does not meet the format requirements, the data to be evaluated can be further modified until the format requirements are met.

[0097] In actual implementation, the index items and index values ​​of the associated data can be directly extracted, and the extracted index items and index values ​​that are not included in the data to be evaluated are added to the data to be evaluated, which can help to make up for the data missing in the data to be evaluated.

[0098] In the above implementation manner, since the format of the data to be evaluated can be checked based on the format requirements before the indicators are determined based on the data to be evaluated, when the data to be evaluated does not meet the format requirements, the related data of the data to be evaluated is retrieved in the target cost consulting data, and the data to be evaluated is corrected based on the related data. This can help to correct the data to be evaluated in combination with the content related to the data to be evaluated in the target cost consulting data when there is missing information in the data to be evaluated, thereby helping to reduce the problem of indicator extraction errors caused by missing information in the data to be evaluated, and further help to improve the accuracy of the investment estimation results.

[0099] In some embodiments, reference Figure 4 Before generating the investment estimation data corresponding to the target project based on the indicator value of the preset indicator item, the following steps are also included: Step 401: for each indicator value, determine whether the indicator value belongs to the corresponding data range.

[0100] The data range is pre-set based on the type of indicator item corresponding to the indicator value. Specifically, the data range can be a numerical interval, such as 0-10000, or a data set, such as the number of various concretes, which can be determined according to the characteristics of the indicator values ​​corresponding to different types of indicator items.

[0101] Furthermore, the indicator value may include numerical values ​​and text. In this case, the numerical values ​​and text can be compared separately, for example, comparing the numerical values ​​with numerical ranges and comparing the text with data sets. This can help to achieve accurate judgment of the indicator value.

[0102] Furthermore, the numbers in the indicator value may be expressed in text form, such as: one thousand square meters, and the data range is expressed in a numerical interval. At this time, the content of the indicator value can be semantically parsed first, and the numbers expressed in text form can be converted into numerical form, such as: converting one thousand square meters into 1000 square meters. At the same time, the unit of the indicator value may also be expressed in symbols, such as: m 2 In order to facilitate the verification of the unit of the indicator value, it is necessary to convert the symbols in the indicator value into corresponding text, for example: 2 Convert to square meters. This makes it easier to compare indicator values.

[0103] Step 402: When the indicator value does not fall within the data range, determine whether to modify the indicator value.

[0104] Among them, the goal of determining whether to revise the indicator value is to determine whether the indicator value can be revised. Specifically, there are many situations that cause the indicator value to be out of the data range, such as: errors in the indicator value extraction process, unit differences in the indicator value, errors in the correspondence between the indicator value and the indicator item, etc. Some of these errors can be solved by correcting the indicator value later, while some errors are difficult to solve by correcting the indicator value, so it is necessary to determine whether it is necessary to correct the indicator value.

[0105] In one example, determining whether to modify an indicator value includes: determining whether the indicator value and the data type corresponding to the data range are the same; if not, determining that the indicator value does not need to be modified, for example: the data range is a numerical interval, and the indicator value does not include a numerical value; if so, determining that the indicator value needs to be modified.

[0106] In another example, determining whether to modify the indicator value includes: determining whether to modify the indicator value based on the type of the indicator item. Whether the indicator values ​​corresponding to different indicator items can be modified is preset, for example, for some key indicator items, in order to ensure the accuracy of their indicator values, automatic modification of their indicator values ​​is not allowed.

[0107] Step 403: When it is determined that the indicator value needs to be corrected, the indicator value is corrected based on the correction method corresponding to the indicator value.

[0108] The correction method is pre-set based on the type of the indicator item corresponding to the indicator value.

[0109] Optionally, when it is determined that the indicator value does not need to be corrected, abnormal prompt information can be generated based on the abnormal indicator value and its corresponding indicator item to prompt the staff to handle the abnormal indicator value.

[0110] In one example, for an indicator value that has a numerical value, the correction method includes converting the unit of the indicator value into a preset unit corresponding to the indicator item, for example: if the indicator item is area, then the preset unit corresponding to the indicator value is square meters. Accordingly, the indicator value is corrected based on the correction method corresponding to the indicator value, including: determining whether the unit of the indicator value is the same as the preset unit corresponding to the indicator item; if not, converting the unit of the indicator value into the preset unit corresponding to the indicator item according to the preset unit conversion rule. For example: if the unit corresponding to the indicator value is hectares, then hectares need to be converted into square meters. It should be noted that this is not a simple unit replacement of the indicator value. As the unit of the indicator value changes, the numerical value corresponding to the indicator value will also change.

[0111] In another example, the indicator value is text, such as safety requirements, environmental protection requirements, etc. At this time, the correction method includes replacing the indicator value with the corresponding data in the data range. Accordingly, the indicator value is corrected based on the correction method corresponding to the indicator value, including: calculating the text and / or semantic similarity between the indicator value and each piece of data in the data range, and replacing the indicator value with the data with the highest similarity. Specifically, the method for determining the data with the highest similarity refers to the method for determining the preset indicator value with the highest similarity to the indicator item to be processed in the implementation corresponding to the above step 104, which will not be repeated here.

[0112] In the above implementation, before generating investment estimation data based on the indicator value, it is possible to first detect whether the indicator value belongs to the corresponding data range. When the indicator value does not belong to the data range and needs to be corrected, the indicator value can be corrected in combination with the correction method corresponding to the indicator value. In this way, the indicator value can be checked and processed for abnormalities before investment estimation, which can help reduce the impact of abnormal indicator values ​​on investment estimation results, and further help improve the accuracy of investment estimation results.

[0113] In some embodiments, reference Figure 5 , step 105, after generating the investment estimation data corresponding to the target project based on the index value of the preset index item, the following steps are also included: Step 501, determining the level interval to which the investment estimation data belongs.

[0114] The level interval is defined based on the investment budget. In one example, the investment interval includes over-budget and under-budget. In actual implementation, the over-budget and under-budget intervals can be further divided, which can help to judge the situation of investment data.

[0115] Step 502: when the level interval indicates that the investment estimation data exceeds the budget, determine the indicator items exceeding the budget from the various indicator items.

[0116] Optionally, the proportion of the indicator item in the total investment is determined in advance. In this case, the budget value corresponding to each indicator item can be determined based on the investment budget, that is, the proportion of each indicator item, and the investment value corresponding to the indicator item can be determined based on the indicator value. Indicator items with investment values ​​greater than the budget values ​​are determined as indicator items that exceed the budget.

[0117] In actual implementation, a comprehensive analysis can also be performed on multiple indicator items, for example: comprehensively calculating whether the sum of the investment values ​​corresponding to the multiple indicator items is greater than the corresponding budget value; when the sum of the investment values ​​is greater than the corresponding budget value, determining that there are abnormal indicator items that exceed the budget among the multiple indicator items, and specifically determining the indicator items that exceed the budget from the multiple indicator items.

[0118] Step 503: determine abnormal indicator items from the indicator items that exceed the budget.

[0119] Optionally, determining abnormal indicator items from the indicator items that exceed the budget includes: for each indicator item that exceeds the budget, determining the difference between the investment value corresponding to the indicator item and the budget value as the excess value corresponding to the indicator item; and determining abnormal indicator items from the indicator items that exceed the budget in combination with the excess value.

[0120] In one example, determining abnormal indicator items from the indicator items exceeding the budget in combination with the excess value includes: sorting the indicator items exceeding the budget from large to small according to the excess value, and determining the indicator items with a preset number of digits (such as the first three digits) before sorting as abnormal indicator items.

[0121] In another example, abnormal indicator items are determined from the indicator items that exceed the budget in combination with the excess value, including: for each indicator item that exceeds the budget, a reference value corresponding to the indicator item is determined based on the product of an adjustment parameter corresponding to the indicator item and the excess value corresponding to the indicator item; the indicator items that exceed the budget are sorted in descending order according to the reference value, and the indicator items in a preset position before sorting are determined as abnormal indicator items.

[0122] Among them, the adjustment parameters of different indicator items are pre-set. In one example, the adjustment parameter is set according to the importance of the indicator item. The higher the importance of the indicator item, the smaller the corresponding adjustment parameter. In another example, the adjustment parameter is set according to the proportion of the indicator item to the total investment. The greater the proportion of the indicator item to the total investment amount, the smaller the corresponding adjustment parameter.

[0123] Step 504: output the abnormal index item and the index value corresponding to the abnormal index item.

[0124] Optionally, outputting the abnormal indicator item and the indicator value corresponding to the abnormal indicator item includes: generating abnormal prompt information including the abnormal indicator item and the indicator value corresponding to the abnormal indicator item, and outputting the abnormal prompt information.

[0125] In the above implementation, after determining the investment estimation data, the level interval to which the investment estimation data belongs can be determined. When the level interval indicates that the investment estimation data exceeds the budget, abnormal indicator items are determined from the indicator items that exceed the budget, and the abnormal indicator items and their corresponding indicator values ​​are output. This can help to determine and output abnormal indicator items and corresponding indicator values ​​in combination with the actual conditions of each indicator when the investment estimation data exceeds the budget, and can further guide the staff to handle the abnormal indicator items accordingly when the investment estimation is abnormal.

[0126] The present application also provides an electronic device. Figure 6 As shown, Figure 6 The electronic device 600 shown includes: a processor 601 and a memory 603. The processor 601 and the memory 603 are connected, such as through a bus 602. Optionally, the electronic device 600 may also include a transceiver 604. It should be noted that in actual applications, the transceiver 604 is not limited to one, and the structure of the electronic device 600 does not constitute a limitation on the embodiments of the present application.

[0127] Processor 601 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It may implement or execute various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 601 may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, and the like.

[0128] The bus 602 may include a path to transmit information between the above components. The bus 602 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus. The bus 602 may be divided into an address bus, a data bus, etc. For ease of representation, Figure 6 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0129] The memory 603 may be a ROM (Read Only Memory) or other types of static storage devices that can store static information and instructions, a RAM (Random Access Memory) or other types of dynamic storage devices that can store information and instructions, or an EEPROM (Electrically Erasable Programmable Read Only Memory), a disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store desired program codes in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto.

[0130] The memory 603 is used to store the application code for executing the solution of the present application, and the execution is controlled by the processor 601. The processor 601 is used to execute the application code stored in the memory 603 to implement the contents shown in the above method embodiment.

[0131] The electronic devices include, but are not limited to, mobile terminals such as mobile phones, laptop computers, PDAs (personal digital assistants), PADs (tablet computers), etc., and fixed terminals such as digital TVs, desktop computers, etc. It can also be a server terminal, etc. Figure 6 The electronic device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0132] An embodiment of the present application also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed in a computer, the computer is caused to execute the method for generating investment estimation data for engineering cost consulting provided in the above embodiment.

[0133] It should be understood that although the steps in the flowchart of the accompanying drawings are shown in sequence according to the instructions of the arrows, these steps are not necessarily performed in the order indicated by the arrows. Unless otherwise clearly stated in this document, the execution of these steps is not strictly limited in order and can be performed in other orders.

[0134] The above are only some implementation methods of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for generating investment estimation data for engineering cost consulting, characterized in that: The method comprises: Obtain the target cost consulting data corresponding to the target project; Embedding a building category feature corresponding to the target project into the target cost consulting data, wherein the building category feature is used to indicate the project category of the target project; Inputting the target cost consulting data into a pre-trained data extraction model to be evaluated, obtaining data to be evaluated corresponding to the target project, wherein the type of the data to be evaluated corresponds to the project category of the target project, and the data extraction model to be evaluated includes a multi-head attention network layer; Determine an index value of a preset index item corresponding to the target project based on the data to be evaluated, wherein the preset index item is related to the project category of the target project; The investment estimation data corresponding to the target project is generated based on the indicator value of the preset indicator item.

2. The method according to claim 1, characterized in that The data extraction model to be evaluated is trained based on the following method: Obtain historical engineering cost consulting data for different preset engineering categories; For each historical engineering cost consulting data, embedding the building category feature corresponding to the historical engineering cost consulting data into the historical engineering cost consulting data, and determining the sample data to be evaluated in the historical cost evaluation data based on the type of data to be evaluated corresponding to the engineering category of the historical engineering cost consulting data; Determine each of the historical engineering cost consulting data and the engineering category and sample data to be evaluated corresponding to the historical engineering cost consulting data as a set of sample data; A training data set is constructed based on the sample data to train the pre-constructed initial network model in batches to obtain the data extraction model to be evaluated, and each of the training data sets includes the sample data of at least two engineering categories.

3. The method according to claim 1, characterized in that The number of attention heads in the multi-head attention network layer matches the number of data types to be evaluated corresponding to each preset engineering category.

4. The method according to claim 3, characterized in that The attention network layer includes a first attention head and a second attention head, wherein the first attention head is used to extract evaluation data types common to at least two of the preset engineering categories; The second attention head is used to extract the evaluation data type used by the specific preset engineering category.

5. The method according to claim 3 or 4, characterized in that: The multi-head attention network layer also includes a category prediction module; The category prediction module is used to predict the engineering category to which the input data belongs, so as to guide the multi-head attention network to adjust the calculation strategy of the attention head based on the working category.

6. The method according to claim 1, characterized in that Before determining the index values ​​of each preset index item of the engineering category of the target engineering based on the data to be evaluated, the method further includes: For each of the data to be evaluated, determining whether the data to be evaluated meets preset format requirements; In the case where the data to be evaluated does not meet the format requirements, searching the target cost consulting data for associated data of the data to be evaluated based on the data to be evaluated; The data to be evaluated is modified based on the associated data.

7. The method according to claim 1, characterized in that Before generating the investment estimation data corresponding to the target project based on the indicator value of the preset indicator item, the method further includes: For each of the indicator values, determining whether the indicator value belongs to a corresponding data range, the data range being pre-set based on a type of an indicator item corresponding to the indicator value; If the indicator value does not fall within the data range, determining whether to modify the indicator value; In the case where it is determined to correct the indicator value, the indicator value is corrected based on a correction method corresponding to the indicator value, and the correction method is pre-set based on a type of an indicator item corresponding to the indicator value.

8. The method according to claim 1, characterized in that After generating the investment estimation data corresponding to the target project, the method further includes: Determining the level interval to which the investment estimation data belongs; When the level interval indicates that the investment estimation data exceeds the budget, determining the indicator item exceeding the budget from each of the indicator items; Determining abnormal indicator items from the indicator items exceeding the budget; Output the abnormal indicator item and the indicator value corresponding to the abnormal indicator item.

9. An electronic device, characterized in that: The electronic device comprises: at least one processor; Memory; At least one application, wherein the at least one application is stored in a memory and configured to be executed by at least one processor, and the at least one application is configured to: execute the investment estimation data generation method for engineering cost consulting as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed in a computer, the computer is caused to execute the method for generating investment estimation data for engineering cost consulting according to any one of claims 1 to 8.

Images