Consulting agency capacity quantification evaluation method based on whole-process transaction operation data

Abstract

The application discloses a consulting agency capacity quantitative evaluation method based on whole-process transaction matching data and belongs to the technical field of data mining. The method comprises the following steps: extracting transaction behavior words, evaluation tendency words and progress deviation words; obtaining a characteristic word set with time identifiers; according to the distance between the time identifiers and the current time, each characteristic word in the characteristic word set with time identifiers is given a decreasing weight to obtain a decay characteristic word set; inputting the decay characteristic word set into a pre-trained capacity evaluation model to obtain an agency capacity quantitative score output by the model; determining an agency capacity level; and generating and outputting an evaluation report. The application completely replaces the traditional self-reporting mode by collecting platform real transaction matching data, constructs a three-dimensional objective evaluation model integrating "transaction performance + service reputation + process performance", and realizes objective, accurate and dynamic quantitative evaluation of the capacity of the consulting agency.

Description

Technical Field

[0001] This invention relates to the field of data mining technology, and in particular to a method for quantitatively evaluating the capabilities of consulting firms based on full-process transaction data. Background Technology

[0002] Full-process engineering consulting refers to integrated consulting services covering all stages of a construction project's lifecycle, including planning and consulting, preliminary feasibility studies, engineering design, bidding agency, cost consulting, engineering supervision, construction process management, final acceptance, and operation and maintenance. With the rapid development of the full-process engineering consulting market and the widespread emergence of various bidding platforms and consulting transaction platforms, how to scientifically, objectively, and dynamically evaluate the true service capabilities of consulting firms has become a critical issue that the industry urgently needs to address. It is also the core technological foundation for platforms to achieve accurate business opportunity matching and intelligent recommendations.

[0003] However, the aforementioned prior art has the following significant shortcomings:

[0004] First, the data source relies on self-reporting by the institutions, making objectivity difficult to guarantee. Patent CN120494734A integrates relevant data from the entire engineering consulting process and uploads it to a computer. This data includes bidding documents, engineering cost data, and construction progress data. This data collection method relies on consulting institutions compiling and uploading data themselves, making it difficult to avoid the subjectivity and manipulation of the data—to obtain higher scores, the reporting parties often exaggerate their achievements or conceal negative records, leading to distorted evaluation results. Existing evaluation systems also lack effective support from large databases for data verification, making it difficult to guarantee the accuracy of the verification.

[0005] Second, the evaluation indicators primarily rely on static data, lacking dynamic measurement of process performance. Existing indicator systems mainly label and classify the static characteristics of bidding documents, engineering cost data, and construction progress data, failing to incorporate the dynamic performance of consulting firms during actual project execution (such as schedule achievement and investment control deviations) into the core evaluation system. These are precisely the core parameters reflecting the consulting firm's true management capabilities. Existing methods can only assess "what conditions are met," but cannot effectively measure "how well they are performed."

[0006] Third, the evaluation process relies on subjective expert judgment, resulting in poor consistency and repeatability. CN120494734A uses expert scoring and the analytic hierarchy process to determine the weights of each indicator. The evaluation process is greatly influenced by the experience and preferences of the expert group, and different expert groups may give significantly different evaluation results, affecting the objectivity and repeatability of the evaluation.

[0007] Fourth, the assessment results are static and lack dynamic updating capabilities. Current assessment technologies are based on single data uploads and single calculations. Once the assessment is completed, it cannot be automatically updated as the organization completes new projects, making it difficult to reflect the dynamic changes in the organization's capabilities.

[0008] Meanwhile, many current bidding and tendering transaction platforms and full-process consulting service platforms have achieved online transaction matching throughout the entire process, accumulating massive amounts of high-value data. This includes transaction records between institutions and project owners, historical quotations, project execution milestones, budget execution rates, and satisfaction and quality ratings given by owners after the transaction. This "transaction matching data" inherently possesses the characteristics of objectivity, authenticity, and real-time performance, making it a valuable asset for evaluating the capabilities of institutions. However, the industry still lacks a method to effectively integrate transaction data, service evaluation data, and process performance data from such platforms, and to conduct precise quantitative evaluations based on this data.

[0009] In summary, existing technologies for assessing the capabilities of full-process engineering consulting firms suffer from several technical shortcomings, including highly subjective data sources, static evaluation indicators, reliance on human judgment in the assessment process, and a lack of dynamic updates to the results. There is an urgent need for a solution that replaces self-reported data with real platform transaction data from the entire process, incorporates process performance indicators into the core evaluation, and achieves automated dynamic assessment. Summary of the Invention

[0010] This application provides a quantitative evaluation method for consulting firm capabilities based on full-process transaction data. By collecting real transaction matching data from the platform (transaction records, service evaluations, historical quotations), it completely replaces the traditional self-reporting method. It incorporates process performance indicators such as project completion rate and investment control deviation rate into the core evaluation system, and constructs a three-dimensional objective evaluation model that integrates "transaction performance + service reputation + process performance". This enables an objective, accurate, and dynamic quantitative evaluation of consulting firm capabilities, providing highly credible data for the platform's intelligent business opportunity recommendation, institution screening, and capability profiling.

[0011] This application provides a method for quantitatively evaluating the capabilities of consulting firms based on full-process transaction operation data, including: obtaining the institution's transaction record text, business evaluation text, and project node record text recorded by the engineering consulting transaction platform, and extracting transaction behavior words, evaluation tendency words, and progress deviation words;

[0012] Based on the data generation time, time markers are added to transaction behavior words, evaluation tendency words, and progress deviation words to obtain a set of feature words with time markers;

[0013] Based on the distance of the time marker from the current time, each feature word in the feature word set with the time marker is assigned a decreasing weight, resulting in a decayed feature word set.

[0014] Input the set of decaying feature words into a pre-trained capability assessment model to obtain the quantitative score of the institution's capability output by the model;

[0015] The organization's capability score is matched with a preset score range to determine the organization's capability level.

[0016] An assessment report is generated and output based on the organization's capability score and capability level.

[0017] Furthermore, the steps to obtain the institutional transaction record text, business evaluation text, and project milestone record text from the engineering consulting transaction platform include:

[0018] Extract the winning bid notification and quotation text from the historical order database of the engineering consulting transaction platform as the institution's transaction record text; extract the satisfaction and quality comments filled in by the client as the business evaluation text; and extract the planned construction period, actual construction period, planned investment amount, and actual investment amount as the project node record text.

[0019] The institutional transaction record text, business evaluation text, and project node record text are compiled to form the original text set;

[0020] The original text set is deduplicated and filtered for stop words to obtain cleaned institutional transaction record text, business evaluation text, and project node record text.

[0021] Further steps to extract transaction behavior terms, evaluation tendency terms, and progress deviation terms include:

[0022] The cleaned institutional transaction record text is segmented using a pre-set dictionary, and numerical terms representing transaction frequency and monetary terms representing price levels are extracted and combined as transaction behavior terms.

[0023] The cleaned business evaluation text is segmented using a pre-set sentiment dictionary, and positive words indicating satisfaction and negative words indicating dissatisfaction are extracted and combined as evaluation sentiment words.

[0024] By comparing the planned construction period text with the actual construction period text, time difference words indicating that the time has exceeded or been completed ahead of schedule are extracted.

[0025] By comparing the text of planned investment amount with the text of actual investment amount, we can extract the difference words that indicate overspending or savings.

[0026] Combine terms related to time difference and funding difference to create terms related to progress deviation.

[0027] Furthermore, the steps to obtain the set of feature words with time identifiers include:

[0028] Identify the generation date of the cleaned institutional transaction record text, the reporting date of the project node record text, and the submission date of the business evaluation text. Convert the generation date, the reporting date, and the submission date of the business evaluation text into standard date format, and use them as time identifiers to append to the beginning of the transaction behavior words, the beginning of the schedule deviation words, and the beginning of the evaluation tendency words, respectively.

[0029] The transaction behavior words, evaluation tendency words, and progress deviation words with time stamps are concatenated in a preset order to form a string consisting of time stamps and word content;

[0030] Store the string in a preset array to obtain a set of feature words with time identifiers.

[0031] Furthermore, the steps to obtain the set of decaying feature words include:

[0032] Extract time identifiers from the set of feature words with time identifiers from a preset array;

[0033] Obtain the current system date as a reference date, and calculate the month difference between the date indicated by the time stamp and the reference date;

[0034] Determine the preset interval within which the difference for that month falls;

[0035] When the difference between months falls within the first interval, the corresponding feature word is assigned the first weight value.

[0036] When the difference between months falls within the second interval, the corresponding feature word is assigned a second weight value.

[0037] The difference between the months corresponding to the first interval is less than the difference between the months corresponding to the second interval;

[0038] The first weight value is greater than the second weight value;

[0039] Each feature word is concatenated with its corresponding first or second weight value to obtain a set of decaying feature words.

[0040] Furthermore, the steps to obtain the quantitative score of the mechanism capability output by the model include:

[0041] Map each feature word in the decay feature word set to a multi-dimensional floating-point vector, and combine the multi-dimensional floating-point vectors into a word vector matrix;

[0042] By inputting the word vector matrix into a pre-trained capability assessment model, transaction feature representation, evaluation feature representation, and progress feature representation are obtained.

[0043] By concatenating the transaction feature representation, evaluation feature representation, and progress feature representation, a comprehensive feature representation is obtained.

[0044] The output layer calculates and outputs the quantitative score of the mechanism capability based on the comprehensive feature representation.

[0045] Further steps in determining an organization's capability level include:

[0046] Obtain the quantitative score of the mechanism capability output by the output layer, and read multiple pre-divided score intervals from the storage medium;

[0047] The quantitative score of the institution's capabilities is compared with multiple score ranges one by one to determine the target score range into which the quantitative score of the institution's capabilities falls.

[0048] Read the mapping table between score ranges and institutional capability levels from the storage medium, and find the institutional capability level corresponding to the target score range in the mapping table;

[0049] The organization's capability level is marked as the final level of the consulting organization to be evaluated, and the organization's capability quantification score and the final level are combined to form the rating result;

[0050] The organizational capability levels include Excellent, Good, Fair, Satisfactory, and Need for Improvement.

[0051] Furthermore, the steps for generating and outputting the evaluation report include:

[0052] Obtain the quantitative score and level of the consulting firm's capabilities from the rating results.

[0053] Extract transaction feature representations, evaluation feature representations, and progress feature representations from the hidden layer;

[0054] Convert the transaction feature representation, evaluation feature representation, and progress feature representation into text descriptions respectively;

[0055] Read the preset report template from the storage medium, fill in the organization's capability quantification score, organization capability level, and text description into the preset report template, and generate an assessment report containing the score, level, and text description;

[0056] The assessment report is converted into a readable format that the front end can recognize, and the readable assessment report is sent to the front end display interface of the engineering consulting transaction platform for display.

[0057] At the same time, the evaluation report in a readable format is stored in the database as a historical record.

[0058] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0059] First, it addresses the issue of data subjectivity. This invention completely replaces the traditional self-reporting method by institutions with real transaction matching data (transaction records, historical quotes, service evaluations, etc.) generated on the platform, eliminating the possibility of data manipulation and falsification at the source. All evaluation data comes from real data automatically recorded by the platform during the transaction process, and cannot be tampered with or selectively reported. The evaluation results are entirely determined by objective market behavior and performance records, greatly improving credibility and solving the inherent defects of existing methods that rely on subjective self-evaluation.

[0060] Second, for the first time, process performance indicators are incorporated into the core evaluation, making the assessment more comprehensive. This invention, for the first time, incorporates process performance indicators such as project completion rate and investment control deviation rate as independent dimensions into the core evaluation system of consulting firm capabilities. This allows the evaluation to move beyond static post-project qualification documents and satisfaction surveys, delving into the process quality of project execution. This truly identifies firms that are "capable of taking on projects and, more importantly, adept at doing them well," significantly enhancing the validity of the evaluation.

[0061] Third, the evaluation process is automated and standardized, eliminating interference from human factors. This invention automatically captures transaction data through a data acquisition module, automatically calculates scores for each dimension using quantitative formulas, and automatically assesses capability levels using preset standards. The entire evaluation process requires no manual intervention. Compared to traditional methods that rely on expert scoring and the analytic hierarchy process (AHP) to determine weights, the evaluation results of this invention have high consistency and repeatability, avoiding evaluation biases caused by differences in subjective judgment among different expert groups.

[0062] Fourth, a time decay and dynamic update mechanism is introduced to achieve continuous tracking. This invention utilizes a time decay mechanism and an incremental update mechanism, allowing the capability index to promptly reflect recent changes in an institution's true performance, avoiding assessment distortion caused by long-term historical performance. The trend identification function can identify institutions with continuously declining capabilities in advance, providing decision support for platform risk management.

[0063] Fifth, the multi-dimensional data source cross-validation ensures high robustness. The three data sources—transaction performance, service reputation, and process performance—are independent yet mutually corroborative. For example, an organization with high service ratings but poor process and schedule control will be objectively revealed, preventing manipulation of a single evaluation system.

[0064] Sixth, the weights are dynamically configurable and adaptable to diverse scenarios. The weights can be automatically adjusted according to different consulting business types (such as full-process project management, cost consulting, and bidding agency) and project characteristics. The evaluation results can accurately match specific project needs, effectively empowering the precise matching and intelligent recommendation of bidding information AI business opportunities, and improving the platform's matching efficiency and project success rate.

[0065] Seventh, fully activate the value of platform data assets. This invention fully explores the transaction records, evaluation feedback, node execution, and other data that have been accumulated but not fully utilized by the trading platform over a long period of time, transforming them into quantifiable capability credit assets, providing a reliable data foundation for building a credit system for the consulting industry. Attached Figure Description

[0066] Figure 1 A flowchart illustrating the method for quantitatively evaluating the capabilities of consulting firms based on full-process transaction data, provided in this application embodiment. Detailed Implementation

[0067] This application provides a method for quantitatively evaluating the capabilities of consulting firms based on full-process transaction data. By collecting real transaction matching data (transaction records, service evaluations, historical quotations) from the platform, it completely replaces the traditional self-reporting method. It incorporates process performance indicators such as project completion rate and investment control deviation rate into the core evaluation system, and constructs a three-dimensional objective evaluation model that integrates "transaction performance + service reputation + process performance". This enables an objective, accurate, and dynamic quantitative evaluation of consulting firm capabilities, providing highly reliable data for the platform's intelligent business opportunity recommendation, institution screening, and capability profiling.

[0068] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0069] like Figure 1 The diagram shows a flowchart of a method for quantitatively evaluating the capabilities of consulting firms based on full-process transaction data provided in this application. The method includes the following steps: obtaining the transaction record text, business evaluation text, and project node record text of the engineering consulting transaction platform; performing word segmentation and part-of-speech tagging on the transaction record text, business evaluation text, and project node record text; and extracting transaction behavior words, evaluation tendency words, and progress deviation words.

[0070] Based on the data generation time, time markers are added to transaction behavior words, evaluation tendency words, and progress deviation words to obtain a set of feature words with time markers;

[0071] Based on the distance of the time marker from the current time, each feature word in the feature word set with the time marker is assigned a decreasing weight, resulting in a decayed feature word set.

[0072] Input the set of decaying feature words into a pre-trained capability assessment model to obtain the quantitative score of the institution's capability output by the model;

[0073] The organization's capability score is matched with a preset score range to determine the organization's capability level.

[0074] Generate and output an assessment report according to the quantified institutional capability score and the institutional capability level.

[0075] Further, the step of acquiring the institutional transaction record text, business evaluation text, and project node record text recorded by the engineering consulting trading platform includes:

[0076] Extract the bid winning notice text and quotation sheet text of the institution from the historical order database of the engineering consulting trading platform as the institutional transaction record text, extract the satisfaction comment text and quality comment text filled in by the client as the business evaluation text, and extract the planned construction period text, actual construction period text, planned investment amount text, and actual investment amount text as the project node record text;

[0077] Summarize the institutional transaction record text, business evaluation text, and project node record text to form an original text set;

[0078] Perform deduplication processing and stop word filtering on the original text set to obtain cleaned institutional transaction record text, business evaluation text, and project node record text.

[0079] In this embodiment, the specific execution logic for acquiring the institutional transaction record text, business evaluation text, and project node record text recorded by the engineering consulting trading platform is as follows: the system connects to different business sub-databases of the platform through a preset data access interface.

[0080] First, from the historical order database, use the structured query language to extract the bid winning notice text field and the quotation sheet detail field of the target consulting institution within the set period, and perform field splicing on these two parts of text data to serve as the institutional transaction record text. Second, from the completion feedback database, extract the satisfaction comment text and quality comment text filled in by the client upon project completion as the business evaluation text, and such texts are usually in the form of short texts. At the same time, from the project management database, extract the planned construction period text, actual construction period text, planned investment amount text, and actual investment amount text of the projects undertaken by the institution as the project node record text.

[0081] After acquiring the above three types of texts, the system summarizes them into the memory to form an original text set. Since the collected texts may contain duplicate entries or meaningless stop words (such as "de", "le", "zai", etc.), the system needs to perform deduplication processing on the original text set (for example, calculate the hash values of the texts and compare to eliminate identical hash values) and stop word filtering (load a preset stop word list for regular matching and replacement), and finally obtain the cleaned institutional transaction record text, business evaluation text, and project node record text, which provides a clean data source for subsequent word segmentation.

[0082] Further steps to extract transaction behavior terms, evaluation tendency terms, and progress deviation terms include:

[0083] The cleaned institutional transaction record text is segmented using a pre-set dictionary, and numerical terms representing transaction frequency and monetary terms representing price levels are extracted and combined as transaction behavior terms.

[0084] The cleaned business evaluation text is segmented using a pre-set sentiment dictionary, and positive words indicating satisfaction and negative words indicating dissatisfaction are extracted and combined as evaluation sentiment words.

[0085] By comparing the planned construction period text with the actual construction period text, time difference words indicating that the time has exceeded or been completed ahead of schedule are extracted.

[0086] By comparing the text of planned investment amount with the text of actual investment amount, we can extract the difference words that indicate overspending or savings.

[0087] Combine terms related to time difference and funding difference to create terms related to progress deviation.

[0088] In this embodiment, the detailed processing logic for segmenting and tagging the cleaned text to extract feature words is as follows: The system uses a dictionary-based segmentation algorithm (such as the forward maximum matching algorithm) to segment the cleaned institutional transaction record text. For the segmented tokens, regular expressions are used to extract numerical words indicating transaction frequency (such as "the 3rd time") and monetary words indicating price level (such as "5 million yuan"). The numerical words and monetary words in the same record are combined to form transaction behavior words.

[0089] For business evaluation texts, the system loads a pre-set sentiment dictionary for the engineering consulting field, performs word-by-word matching on the text, extracts positive words indicating satisfaction (such as "professional" and "timely") and negative words indicating dissatisfaction (such as "delayed" and "rough"), and combines them as evaluation sentiment words.

[0090] For project node record text, the system executes text comparison logic: subtract the value in the planned duration text from the value in the actual duration text. If the difference is greater than zero, a time difference term "overdue by X days" is generated; if it is less than or equal to zero, a time difference term "on schedule" or "X days ahead of schedule" is generated. Similarly, the actual investment amount text is subtracted from the planned investment amount text to generate a funding difference term "overspending by X yuan" or "saving X yuan".

[0091] Finally, the time difference words and funding difference words are concatenated into strings to form the progress deviation words. This pure text parsing and numerical comparison logic avoids complex image recognition or deep model extraction.

[0092] Furthermore, the steps to obtain the set of feature words with time identifiers include:

[0093] Identify the generation date of the cleaned institutional transaction record text, the reporting date of the project node record text, and the submission date of the business evaluation text. Convert the generation date, the reporting date, and the submission date of the business evaluation text into standard date format, and use them as time identifiers to append to the beginning of the transaction behavior words, the beginning of the schedule deviation words, and the beginning of the evaluation tendency words, respectively.

[0094] The transaction behavior words, evaluation tendency words, and progress deviation words with time stamps are concatenated in a preset order to form a string consisting of time stamps and word content;

[0095] Store the string in a preset array to obtain a set of feature words with time identifiers.

[0096] In this embodiment, the execution details of constructing a set of feature words with time identifiers are as follows: When parsing the original text, the system synchronously reads the generation timestamp of each record in the database. For institutional transaction record text, the system identifies the generation date of its entry into the database and calls the time formatting function to convert the date into a standard date format string of "YYYYMMDD" as a time identifier. This string is then appended to the beginning of the transaction behavior words through string concatenation, for example, forming the format "20240510500 million yuan".

[0097] Similarly, the system identifies the client's submission date in the business evaluation text, converts it to a standard date format, and appends it to the beginning of the evaluation tendency words; it also identifies the node reporting date in the project node record text, converts it to a standard date format, and appends it to the beginning of the schedule deviation words. Subsequently, the system sorts and concatenates the feature words with time identifiers according to a preset arrangement order (e.g., schedule deviation words first, then evaluation tendency words, and finally transaction behavior words), forming a continuous string composed of time identifiers and word content.

[0098] The system initializes an array object in memory and stores the concatenated strings sequentially into this array, thus obtaining a structured set of time-stamped feature words. This step, through unified time stamp formatting, provides a standardized data input basis for subsequent calculation of time decay weights.

[0099] Furthermore, the steps to obtain the set of decaying feature words include:

[0100] Extract time identifiers from the set of feature words with time identifiers from a preset array;

[0101] Obtain the current system date as a reference date, and calculate the month difference between the date indicated by the time stamp and the reference date;

[0102] Determine the preset interval within which the difference for that month falls;

[0103] When the difference between months falls within the first interval, the corresponding feature word is assigned the first weight value.

[0104] When the difference between months falls within the second interval, the corresponding feature word is assigned a second weight value.

[0105] The difference between the months corresponding to the first interval is less than the difference between the months corresponding to the second interval;

[0106] The first weight value is greater than the second weight value;

[0107] Each feature word is concatenated with its corresponding first or second weight value to obtain a set of decaying feature words.

[0108] In this embodiment, the detailed calculation logic for obtaining the attenuation feature word set is as follows: The system iterates through the array in memory to extract each time identifier from the feature word set with time identifiers, and obtains the current date of the system as the reference date. For each time identifier, the system calculates the month difference between the date it indicates and the reference date. The formula for calculating this month difference can be expressed as:

[0109] ;

[0110] in For reference date year, The year as a time identifier, For the reference date, the month, The month is a time identifier. This refers to the difference in months, and the formula has uniform dimensions and is simple to calculate. The system has preset multiple difference ranges, for example, the first range is 0 to 3 months, the second range is 4 to 6 months, and the third range is 7 to 12 months.

[0111] Determine which interval the calculated monthly difference falls into: When the monthly difference is in the first interval, assign the corresponding feature word a first weight value (e.g., 1.0); when the monthly difference is in the second interval, assign the corresponding feature word a second weight value (e.g., 0.6); when the monthly difference is in the third interval or larger, assign the corresponding feature word a third weight value (e.g., 0.3), where the first weight value is greater than the second weight value, and the second weight value is greater than the third weight value.

[0112] Finally, the system concatenates each feature word with its corresponding weight value, summing all the concatenated strings to obtain the decayed feature word set. This design, through segmented weight assignment, reserves flexible space for interval division and weight value adjustment.

[0113] Furthermore, the steps to obtain the quantitative score of the mechanism capability output by the model include:

[0114] Map each feature word in the decay feature word set to a multi-dimensional floating-point vector, and combine the multi-dimensional floating-point vectors into a word vector matrix;

[0115] The word vector matrix is ​​input into a pre-trained capability assessment model, which includes an input layer, a hidden layer, and an output layer. The input layer receives the word vector matrix and passes it to the hidden layer. The hidden layer extracts features from the word vector matrix for transaction behavior, evaluation tendency, and progress deviation, respectively, to obtain transaction feature representation, evaluation feature representation, and progress feature representation.

[0116] The hidden layer concatenates the transaction feature representation, evaluation feature representation, and progress feature representation to obtain a comprehensive feature representation;

[0117] The output layer calculates and outputs the quantitative score of the mechanism capability based on the comprehensive feature representation.

[0118] In this embodiment, the internal data flow logic for obtaining the quantitative score of institutional capability is as follows: The system first calls a word vector mapping table (such as a mapping dictionary generated by training the Word2Vec model) to map each feature word in the decay feature word set to a fixed-dimensional multi-dimensional floating-point vector (e.g., 50-dimensional). The system arranges all word vectors in order and combines them into a word vector matrix. This word vector matrix is ​​input into a pre-trained capability assessment model. The capability assessment model is a lightweight multilayer perceptron network structure, including an input layer, a hidden layer, and an output layer. The input layer receives the word vector matrix and linearly passes it to the hidden layer. The hidden layer has three parallel feature processing branches designed to extract features and perform weighted summation on the transaction behavior word vectors, evaluation tendency word vectors, and progress deviation word vectors in the word vector matrix, respectively, to obtain transaction feature representation, evaluation feature representation, and progress feature representation.

[0119] Subsequently, the hidden layer concatenates these three types of feature representations along the feature dimension to obtain a long vector of comprehensive feature representation. The output layer receives this comprehensive feature representation, maps it to the 0-1 range through a fully connected layer and an activation function (such as the sigmoid function, then multiplies it by 100), calculates and outputs a floating-point number, which is the quantification score of the organization's capability. This step achieves feature fusion and scoring through a simple fully connected network structure, ensuring the effectiveness of the model while avoiding the computational limitations caused by using an overly large model.

[0120] Further steps in determining an organization's capability level include:

[0121] Obtain the quantitative score of the mechanism capability output by the output layer, and read multiple pre-divided score intervals from the storage medium;

[0122] The quantitative score of the institution's capabilities is compared with multiple score ranges one by one to determine the target score range into which the quantitative score of the institution's capabilities falls.

[0123] Read the mapping table between score ranges and institutional capability levels from the storage medium, and find the institutional capability level corresponding to the target score range in the mapping table;

[0124] The organization's capability level is marked as the final level of the consulting organization to be evaluated, and the organization's capability quantification score and the final level are combined to form the rating result;

[0125] The organizational capability levels include Excellent, Good, Fair, Satisfactory, and Need for Improvement.

[0126] In this embodiment, the operational details for determining the organizational capability level are as follows: After obtaining the quantitative score of organizational capability output by the output layer, the system reads multiple pre-divided score interval configuration files from the local storage medium or configuration center. These configuration files define multiple consecutive score intervals. The system compares the quantitative score of organizational capability with each of the multiple score intervals in descending order, determining whether the score is greater than or equal to the lower limit of the current interval and less than the upper limit of the current interval, thereby determining the target score interval into which the quantitative score of organizational capability falls.

[0127] Simultaneously, the system reads a mapping table between score ranges and institutional capability levels from the storage medium. This mapping table records the correspondence between each score range and the institutional capability level. The system then finds the institutional capability level corresponding to the target score range in the mapping table and marks that institutional capability level as the final level of the consulting firm to be evaluated.

[0128] The organization's capability level is clearly divided into five levels: Excellent, Good, Fair, Satisfactory, and Needs Improvement. The system combines the organization's capability quantification score and final level into a single rating result record, which is stored in the result cache for subsequent use by the report generation module.

[0129] Furthermore, the steps for generating and outputting the evaluation report include:

[0130] Obtain the quantitative score and level of the consulting firm's capabilities from the rating results.

[0131] Extract transaction feature representations, evaluation feature representations, and progress feature representations from the hidden layer;

[0132] Convert the transaction feature representation, evaluation feature representation, and progress feature representation into text descriptions respectively;

[0133] Read the preset report template from the storage medium, fill in the organization's capability quantification score, organization capability level, and text description into the preset report template, and generate an assessment report containing the score, level, and text description;

[0134] The assessment report is converted into a readable format that the front end can recognize, and the readable assessment report is sent to the front end display interface of the engineering consulting transaction platform for display.

[0135] At the same time, the evaluation report in a readable format is stored in the database as a historical record.

[0136] In this embodiment, the specific execution method for generating and outputting the evaluation report is as follows: The system retrieves the rating results of the consulting firm to be evaluated from the result cache, and extracts the quantitative score and level of the firm's capabilities. Simultaneously, the system backtracks to the hidden layer of the capability assessment model, extracting the transaction feature representation, evaluation feature representation, and progress feature representation generated during the feature extraction process. The system calls preset label mapping rules to convert these three types of multi-dimensional feature representations into human-readable text descriptions, for example, converting the high-weight vector in the progress feature representation into the text description "excellent project progress control capability."

[0137] Next, the system reads a preset report template from the storage medium, which defines the paragraph structure and placeholders of the report. The system then fills in the corresponding placeholders in the preset report template with the organization's capability quantification score, capability level, and generated text descriptions through string replacement, generating a complete assessment report containing the score, level, and text descriptions.

[0138] Subsequently, the system converts the assessment report into a readable format recognizable by the front end (such as JSON or HTML format), and sends the readable assessment report to the front-end display interface of the engineering consulting transaction platform for visualization through the application programming interface. At the same time, the readable assessment report is persistently stored in the database as a historical record for later review.

[0139] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0140] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0141] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0142] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0143] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0144] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for quantitatively evaluating the capabilities of consulting firms based on comprehensive transaction data, characterized in that: Includes the following steps: Obtain the institutional transaction record text, business evaluation text, and project node record text from the engineering consulting transaction platform, and extract transaction behavior words, evaluation tendency words, and schedule deviation words; Based on the data generation time, time markers are added to transaction behavior words, evaluation tendency words, and progress deviation words to obtain a set of feature words with time markers; Based on the distance of the time marker from the current time, each feature word in the feature word set with the time marker is assigned a decreasing weight, resulting in a decayed feature word set. Input the set of decaying feature words into a pre-trained capability assessment model to obtain the quantitative score of the institution's capability output by the model; The organization's capability score is matched with a preset score range to determine the organization's capability level. An assessment report is generated and output based on the organization's capability score and capability level.

2. The method for quantitatively evaluating the capabilities of consulting firms based on full-process transaction data as described in claim 1, characterized in that, The steps to obtain the institutional transaction record text, business evaluation text, and project milestone record text from the engineering consulting transaction platform include: Extract the winning bid notification and quotation text from the historical order database of the engineering consulting transaction platform as the institution's transaction record text; extract the satisfaction and quality comments filled in by the client as the business evaluation text; and extract the planned construction period, actual construction period, planned investment amount, and actual investment amount as the project node record text. The institutional transaction record text, business evaluation text, and project node record text are compiled to form the original text set; The original text set is deduplicated and filtered for stop words to obtain cleaned institutional transaction record text, business evaluation text, and project node record text.

3. The method for quantitatively evaluating the capabilities of consulting firms based on full-process transaction data as described in claim 1, characterized in that, The steps for extracting transaction behavior terms, evaluation tendency terms, and progress deviation terms include: The cleaned institutional transaction record text is segmented using a pre-set dictionary, and numerical terms representing transaction frequency and monetary terms representing price levels are extracted and combined as transaction behavior terms. The cleaned business evaluation text is segmented using a pre-set sentiment dictionary, and positive words indicating satisfaction and negative words indicating dissatisfaction are extracted and combined as evaluation sentiment words. By comparing the planned construction period text with the actual construction period text, time difference words indicating that the time has exceeded or been completed ahead of schedule are extracted. By comparing the text of planned investment amount with the text of actual investment amount, we can extract the difference words that indicate overspending or savings. Combine terms related to time difference and funding difference to create terms related to progress deviation.

4. The method for quantitatively evaluating the capabilities of consulting firms based on full-process transaction data as described in claim 1, characterized in that, The steps to obtain the set of feature words with time identifiers include: Identify the generation date of the cleaned institutional transaction record text, the reporting date of the project node record text, and the submission date of the business evaluation text. Convert the generation date, the reporting date, and the submission date of the business evaluation text into standard date format, and use them as time identifiers to append to the beginning of the transaction behavior words, the beginning of the schedule deviation words, and the beginning of the evaluation tendency words, respectively. The transaction behavior words, evaluation tendency words, and progress deviation words with time stamps are concatenated in a preset order to form a string consisting of time stamps and word content; Store the string in a preset array to obtain a set of feature words with time identifiers.

5. The method for quantitatively evaluating the capabilities of consulting firms based on full-process transaction data as described in claim 1, characterized in that, The steps to obtain the set of decaying feature words include: Extract time identifiers from the set of feature words with time identifiers from a preset array; Obtain the current system date as a reference date, and calculate the month difference between the date indicated by the time stamp and the reference date; Determine the preset interval within which the difference for that month falls; When the difference between months falls within the first interval, the corresponding feature word is assigned the first weight value. When the difference between months falls within the second interval, the corresponding feature word is assigned a second weight value. The difference between the months corresponding to the first interval is less than the difference between the months corresponding to the second interval; The first weight value is greater than the second weight value; Each feature word is concatenated with its corresponding first or second weight value to obtain a set of decaying feature words.

6. The method for quantitatively evaluating the capabilities of consulting firms based on full-process transaction data as described in claim 1, characterized in that, The steps to obtain the mechanism capability quantification score output by the model include: Map each feature word in the decay feature word set to a multi-dimensional floating-point vector, and combine the multi-dimensional floating-point vectors into a word vector matrix; By inputting the word vector matrix into a pre-trained capability assessment model, transaction feature representation, evaluation feature representation, and progress feature representation are obtained. By concatenating the transaction feature representation, evaluation feature representation, and progress feature representation, a comprehensive feature representation is obtained. The output layer calculates and outputs the quantitative score of the mechanism capability based on the comprehensive feature representation.

7. The method for quantitatively evaluating the capabilities of consulting firms based on full-process transaction data as described in claim 1, characterized in that, The steps to determine an organization's capability level include: Obtain the quantitative score of the mechanism capability output by the output layer, and read multiple pre-divided score intervals from the storage medium; The quantitative score of the institution's capabilities is compared with multiple score ranges one by one to determine the target score range into which the quantitative score of the institution's capabilities falls. Read the mapping table between score ranges and institutional capability levels from the storage medium, and find the institutional capability level corresponding to the target score range in the mapping table; The organization's capability level is marked as the final level of the consulting organization to be evaluated, and the organization's capability quantification score and the final level are combined to form the rating result; The organizational capability levels include Excellent, Good, Fair, Satisfactory, and Need for Improvement.

8. The method for quantitatively evaluating the capabilities of consulting firms based on full-process transaction data as described in claim 1, characterized in that, The steps to generate and output an evaluation report include: Obtain the quantitative score and level of the consulting firm's capabilities from the rating results. Extract transaction feature representations, evaluation feature representations, and progress feature representations from the hidden layer; Convert the transaction feature representation, evaluation feature representation, and progress feature representation into text descriptions respectively; Read the preset report template from the storage medium, fill in the organization's capability quantification score, organization capability level, and text description into the preset report template, and generate an assessment report containing the score, level, and text description; The assessment report is converted into a readable format that the front end can recognize, and the readable assessment report is sent to the front end display interface of the engineering consulting transaction platform for display. At the same time, the evaluation report in a readable format is stored in the database as a historical record.

Images