Bidding document intelligent generation method based on natural language processing and computer system

Abstract

The invention provides an intelligent bidding document generation method based on natural language processing and a computer system, and the method comprises the steps: dividing a target bidding document into a target text sequence set when the bidding possibility of the target bidding document is analyzed; performing repeated iteration on a first implicit representation set corresponding to the target text sequence set according to a feature accumulator so as to perform cyclic feature extraction on the target text sequence set to obtain a second implicit representation set; the propagation of the core bid invitation semantic information among the text sequences can be completed based on the feature accumulator, so that the second implicit representation set can accurately reflect the core bid invitation semantic information in the target bid invitation file, the bid invitation possibility analysis reliability can be improved when bid invitation possibility analysis is carried out according to the second implicit representation set, and the bid invitation probability analysis efficiency is improved. And an accurate reference is provided for subsequent generation of a suitable bidding file.

Description

Technical Field

[0001] The present application relates to the field of electrical data processing, and more specifically, to a method and computer system for intelligently generating bidding documents based on natural language processing. Background Art

[0002] In modern business activities, bidding is a widely used competition mechanism, which is of vital importance to the development of various projects. In the bidding process, the quality of the bidding documents is directly related to whether the bidder can gain an advantage in the competition. Traditional bidding document preparation mainly relies on manual operation, which involves a lot of text reading, analysis, writing and sorting, consuming a lot of manpower and time.

[0003] Tender documents usually have complex structures and semantic contents, and the tender intent contained in them may not always be intuitive and obvious. When manually interpreted, it is easy to have inaccurate understanding of the tender intent due to differences in individual understanding and different levels of mastery of industry knowledge and project background.

[0004] Before preparing the bidding documents, enterprises need to evaluate the feasibility of the bid, including whether their own capabilities meet the bidding requirements, whether the project risks are controllable, and whether the expected benefits are reasonable. Traditional evaluation methods are often based on empirical judgment or simple cost-benefit analysis, lacking a systematic and comprehensive evaluation method. With the development of information technology, natural language processing (NLP) technology has been widely used in many fields, such as intelligent customer service, information retrieval, and automatic text summarization. However, in the field of bidding, the application of natural language processing technology is relatively small and has not yet fully realized its potential. Summary of the invention

[0005] The purpose of the present invention is to provide a bidding document intelligent generation method and computer system based on natural language processing. This application is implemented as follows: In a first aspect, the present application provides a method for intelligently generating bidding documents based on natural language processing, the method comprising: performing implicit representation extraction on a target text sequence set of a target bidding document to obtain a first implicit representation set; walking the first implicit representation set, and performing the following operation on the xth first implicit representation walked to, wherein x is a positive integer less than X, and X is the number of target text sequences in the target text sequence set: iterating the yth feature accumulator according to the xth first implicit representation to obtain the xth feature accumulator, if x is 1, the yth feature accumulator is the initial feature accumulator, if x is greater than 1, the yth feature accumulator is the feature accumulator of the yth target text sequence, wherein y=x-1; obtaining the xth second implicit representation according to the xth feature accumulator; obtaining the second implicit representation set according to the x second implicit representations obtained by walking the first implicit representation set; performing bidding possibility analysis according to the second implicit representation set to obtain a target bidding support rate; based on the target bidding support rate, calling a matching preset bidding document generation framework to generate a bidding document; In a second aspect, the present application provides a computer system comprising: one or more processors; a memory; one or more computer programs; wherein the one or more computer programs are stored in the memory and configured to be executed by the one or more processors, and when the one or more computer programs are executed by the processors, the method as described above is implemented.

[0006] Beneficial effects of the present application: When performing bidding possibility analysis on a target bidding document, the embodiment of the present application divides the target bidding document into a target text sequence set, and repeatedly iterates the first implicit representation set corresponding to the target text sequence set based on a feature accumulator to perform cyclic feature extraction on the target text sequence set to obtain a second implicit representation set; because the propagation of core bidding semantic information between text sequences can be completed based on the feature accumulator, the second implicit representation set can accurately reflect the core bidding semantic information in the target bidding document, and when performing bidding possibility analysis based on the second implicit representation set, the reliability of the bidding possibility analysis can be increased, providing an accurate reference for the subsequent generation of suitable bidding documents. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 It is a flow chart of a method for intelligently generating bidding documents based on natural language processing provided in an embodiment of the present application; Figure 2 It is a schematic diagram of the composition of a computer system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0008] The execution subject of the bidding document intelligent generation method based on natural language processing in the embodiment of the present application is a computer system, including but not limited to a server, a personal computer, a laptop, a tablet computer, a smart phone, etc. Figure 1 As shown, the method includes: Step S100: performing implicit representation extraction on a target text sequence set of a target bidding document to obtain a first implicit representation set.

[0009] In step S100, the target bidding document is a complete document content, which contains various detailed information about the bidding project, such as the basic requirements, technical specifications, business terms, etc. of the bidding project. The computer system processes the bidding document according to a certain logic or format and converts it into a target text sequence set. The target text sequence set here can be understood as a set of multiple text sequences into which the text in the bidding document is divided according to specific rules. For example, taking a construction project bidding as an example, it may be divided into different text sequences according to different chapters (such as project overview, construction requirements, quality standards, etc.).

[0010] Implicit representation extraction is a process of converting semantic information in a text sequence into a vector representation that can be processed by a computer. Computer systems can use a variety of technical means to achieve implicit representation extraction. A common method is the word vector model, such as the Word2Vec model. Suppose we have a simple text sequence "Concrete strength needs to reach C30". In the Word2Vec model, each word (such as "concrete", "strength", "reach", "C30") will be mapped to a specific vector. The combination of these vectors or further processing (such as weighted summation, averaging, etc.) can obtain the implicit representation of this text sequence. Specifically, if the word vectors are , , , , then a simple implicit representation formula for text sequence can be , where I is the implicit representation of this text sequence.

[0011] The computer system performs such an implicit representation extraction operation on each text sequence in the target text sequence set. For example, for another text sequence in the target text sequence set, "the construction period is 180 days", its implicit representation is obtained in the same manner as described above (using the word vector model and related combination operations). By performing such processing on all text sequences in the target text sequence set, the computer system finally obtains a first implicit representation set. Each element in this first implicit representation set corresponds to an implicit representation of a text sequence in the target text sequence set, which contains a digital representation of the semantic information of each text sequence, and this representation can be used by subsequent operations of the computer system (such as the walking operation in step S200, etc.), so as to further analyze and process the target bidding documents, such as performing bidding possibility analysis and final bidding document generation operations.

[0012] Step S200: Walk the first implicit representation set, and perform the following operation on the x-th first implicit representation walked to, where x is a positive integer less than X, and X is the number of target text sequences in the target text sequence set: iterate the y-th feature accumulator according to the x-th first implicit representation to obtain the x-th feature accumulator, if x is 1, then the y-th feature accumulator is the initial feature accumulator, if x is greater than 1, then the y-th feature accumulator is the feature accumulator of the y-th target text sequence, where y=x-1.

[0013] In step S200, the computer system performs a walk operation on the first implicit representation set obtained in step S100, and performs a specific operation on the xth first implicit representation walked to, where x is a positive integer less than X, and X represents the number of target text sequences in the target text sequence set.

[0014] The so-called walk means that the computer system visits the elements in the first implicit representation set in a certain order. For example, if there are 5 elements in the first implicit representation set (i.e. X=5), the computer system will start from the first element and gradually move to the subsequent elements. This process is like looking at each element in order in an ordered list.

[0015] For the x-th first implicit representation that is walked to, the computer system must iterate the y-th feature accumulator (which can be understood as Hidden State) based on it, and then obtain the x-th feature accumulator. The feature accumulator here is a structure used to store and transmit information. When x is 1, the y-th feature accumulator is the initial feature accumulator, which is the starting state of the entire process. The initial feature accumulator can be initialized to an all-zero vector or a random vector, depending on the technical solution adopted. For example, if an all-zero vector is used for initialization, assuming that the dimension of the feature accumulator is d-dimensional, then the initial feature accumulator H_0 is a d-dimensional vector in which each element is 0, that is (a total of d 0s).

[0016] When x is greater than 1, the yth feature accumulator is the feature accumulator of the yth target text sequence, where y=x-1. This means that during the walk, the computer system will use the feature accumulator information corresponding to the previous text sequence to update the feature accumulator corresponding to the current text sequence. In order to implement this iterative process, the computer system can use a variety of technical means. A common way is to use a combination of linear transformation and nonlinear activation function. Assume that the xth first implicit representation is , the y-th feature accumulator is , the computer system first combines them linearly, e.g. ,in and is a learnable weight matrix whose dimensions are based on and Then, H' is processed by a nonlinear activation function (such as ReLU function) to obtain the xth feature accumulator ,Right now .

[0017] This method of updating the current feature accumulator based on the previous feature accumulator enables the feature accumulator to transfer information between text sequences. Taking a project bidding as an example, assume that there are two different target text sequences about the project budget and the project quality requirements. When the computer system moves from the text sequence of the project budget to the text sequence of the project quality requirements, the information contained in the feature accumulator corresponding to the project budget text sequence (such as the approximate range of the budget, the characteristics of fund allocation, etc.) can be used to update the feature accumulator corresponding to the project quality requirements text sequence, so that the feature accumulator corresponding to the project quality requirements text sequence not only contains its own information (such as quality standards, inspection methods, etc.), but also integrates the relevant information of the previous project budget text sequence. Such an operation helps to integrate information from different parts during the analysis process of the entire target bidding document, laying the foundation for subsequent accurate bidding possibility analysis and bidding document generation.

[0018] By performing such operations on each element in the first implicit representation set (i.e., each first implicit representation), the computer system can gradually construct feature accumulators corresponding to each target text sequence. These feature accumulators will play an important role in subsequent steps (such as obtaining the second implicit representation in step S300), and will help to deeply explore the semantic information and logical relationships in the target bidding documents.

[0019] Step S300: Obtain an xth second implicit representation according to the xth feature accumulator.

[0020] In step S300, the xth feature accumulator is the result obtained by the computer system after a series of operations in step S200. The feature accumulator contains relevant information of the previous text sequence, which, after integration and processing, becomes a state representation that can reflect some features of the target bidding document.

[0021] The computer system needs to obtain the second implicit representation from this feature accumulator. This process can be achieved through a variety of technical means. One possible way is to use linear mapping technology. For example, suppose the xth feature accumulator is a vector with a dimension of d , the computer system can define a weight matrix W with a dimension of e×d (where e is the target dimension of the second implicit representation). Then, through matrix multiplication, the xth second implicit representation can be obtained. , the formula is .

[0022] Take a software project bidding as an example. The target bidding document includes different text sequences about project functional requirements, technical architecture requirements, etc. When the computer system processes the xth feature accumulator corresponding to the technical architecture requirements, this feature accumulator has already integrated the information related to the previous project functional requirements text sequence. The computer system converts this feature accumulator into the xth second implicit representation through the above-mentioned linear mapping operation. This second implicit representation can reflect the technical architecture requirements and the related project functional requirements information in a new form that is more suitable for subsequent analysis.

[0023] Another technical approach can be based on the fully connected layer in the neural network. The xth feature accumulator is used as the input of the fully connected layer, and after calculation by the neurons in the fully connected layer, the xth second implicit representation is output. Each neuron in the fully connected layer is connected to all neurons in the input layer (i.e., the feature accumulator). Each connection has a corresponding weight. The neuron performs weighted summation based on the input value and these weights, and processes it through an activation function (such as the Sigmoid function or the Tanh function) to obtain the final output, which is the xth second implicit representation. This method can process the information in the feature accumulator more flexibly, and mine the complex relationship between different features, thereby generating a more representative second implicit representation, and providing an effective data basis for subsequent operations such as bidding possibility analysis based on the second implicit representation set.

[0024] Step S400: Obtain a second implicit representation set according to x second implicit representations obtained by walking the first implicit representation set.

[0025] In the previous steps, the computer system performs a series of processing on the target text sequence set of the target bidding document. First, in step S100, the first implicit representation set is extracted, and then in step S200, the first implicit representation set is walked and each corresponding feature accumulator is obtained by iterating the feature accumulator, and then in step S300, the second implicit representations are obtained one by one according to these feature accumulators. The second implicit representation here is a new representation form after complex processing of the semantic information of different parts of the target bidding document.

[0026] When the computer system walks through the first implicit representation set, each time it obtains a second implicit representation, it is equivalent to re-encoding and integrating part of the information in the target bidding document. For example, in a construction project bidding, the target bidding document contains multiple aspects such as project scale, construction technology requirements, and construction period requirements. For the text sequence of the project scale part, a specific second implicit representation is obtained through the previous steps. This representation may be a new vector form that integrates information such as the building area and building structure type in the project scale; for the text sequence of the construction technology requirements part, there is also a corresponding second implicit representation, which integrates information such as construction technology and material selection standards.

[0027] As the walking process progresses, the computer system obtains x such second implicit representations. In order to obtain a set of second implicit representations, the computer system can use a simple set construction technique, that is, to combine these x second implicit representations together to form a set. From a mathematical point of view, if we denote the i-th second implicit representation as (where i = 1, 2, ..., x), then the second implicit representation set .

[0028] This method of combining multiple second implicit representations into a set enables the computer system to manage and subsequently process the information after the re-representation of each part of the target bidding document in a holistic form. This second implicit representation set contains rich semantic information of the target bidding document, which has been processed in a form that is more conducive to analysis, and provides a comprehensive data basis for the subsequent bidding possibility analysis (such as step S500) by the computer system, which can reflect the various requirements, conditions and other factors of the target bidding document from a holistic perspective, thereby helping to accurately determine whether to bid and how to generate a suitable bidding document.

[0029] Step S500: Perform bidding possibility analysis based on the second implicit representation set to obtain a target bidding support rate.

[0030] In step S500, the second implicit representation set is the result obtained after processing the target bidding document through the previous series of steps, and it contains the representation forms of various information in the target bidding document after complex transformation. This information covers multiple aspects of the bidding project, such as project requirements, technical specifications, commercial terms, etc.

[0031] Bidding feasibility analysis is a complex process. The computer system mines information related to bidding feasibility from the second implicit representation set. One possible technical approach is to build an evaluation model. For example, a model based on probability statistics can be used. Assume that each element in the second implicit representation set (i.e., each second implicit representation) is a vector (i=1,2,…,n, n is the number of elements in the set), the computer system can define a function To represent the bidding possibility score corresponding to each vector. This function It can be predefined based on historical bidding data. For example, for a specific type of bidding project, if a dimension in a vector (corresponding to a specific bidding requirement) meets specific conditions, a certain score will be given.

[0032] Taking an information technology project bidding as an example, a vector in the second implicit representation set may represent the information of the technical requirements of the project. If the value of an element in the vector indicates that the required technology is highly matched with the technology that the bidder is good at, then according to This function will give a higher score, indicating a higher possibility of bidding in terms of technology.

[0033] The computer system performs this analysis on each element in the second implicit representation set and obtains a series of scores. Then, in order to obtain the target bid support rate, the computer system can use a weighted average method. Assume that each score The corresponding weight is , then the target bid support rate P can be expressed by the formula These weights are calculated. It can be determined based on the importance of different bidding requirements. For example, the weight corresponding to the core technical requirements may be higher, while the weight corresponding to some auxiliary terms may be lower.

[0034] Through such bidding possibility analysis, the target bidding support rate can comprehensively reflect the feasibility of bidding based on the target bidding documents. This ratio will provide an important reference for whether to generate bidding documents and how to generate bidding documents in the future. If the target bidding support rate is high, it means that the possibility of successful bidding is high, so it is more likely to generate bidding documents according to the appropriate framework in the subsequent steps; on the contrary, if the target bidding support rate is low, it is necessary to re-evaluate the necessity of bidding or adjust the bidding strategy.

[0035] Step S600: Based on the target bid support rate, a matching preset bid document generation framework is called to generate a bid document.

[0036] The target bid support rate is a quantitative indicator obtained by in-depth analysis of the target bidding document related information in step S500, which reflects the possibility of successful bidding.

[0037] Preset bid document generation frameworks are a series of pre-built templates or structures designed for different bidding possibilities or project types. For example, in a construction project bidding, if it is a large infrastructure construction project, there may be a preset bid document generation framework specifically for the scale and complexity of such projects; and for small building maintenance projects, there is another framework that is suitable for it. These frameworks contain the structure and format requirements of various parts that bid documents usually need to cover, such as project overview, solution, quotation strategy, project schedule, etc.

[0038] The computer system determines which preset bid document generation framework to call based on the numerical range of the target bid support rate. If the target bid support rate is high, indicating a greater chance of successful bidding, the computer system may call a more comprehensive, detailed and competitive preset bid document generation framework. For example, assuming the target bid support rate exceeds 80%, the framework called by the computer system may elaborate on the advanced technology and unique project management methods used in the solution section, and provide a more advantageous price structure in the quotation strategy section.

[0039] For example, such matching can be achieved by establishing a mapping relationship table. This mapping relationship table records the correspondence between different target bid support rate ranges and the preset bid document generation framework. For example: 80%-100% corresponds to the comprehensive competitiveness framework of large projects; 60%-80% corresponds to the standard framework of medium-sized projects; 40%-60% corresponds to the basic framework of small projects; 0%-40% corresponds to the simple framework for special situations (such as only providing basic information for archiving, etc.).

[0040] After the computer system determines the matching preset tender document generation framework, it starts to generate the tender document. One way is to use a rule-based filling method. Taking the standard framework for medium-sized projects as an example, the project overview part of this framework may define rules that need to include information such as project name, tender unit, and project background. The computer system obtains these contents from the relevant information extracted when processing the target tender document before, and fills them into the corresponding position of the tender document according to the prescribed format. For the solution part, the framework may stipulate that the corresponding technical solutions, human resources arrangements, etc. should be listed according to the project requirements. The computer system generates corresponding content according to the rules based on its own knowledge base or the analysis results of similar projects before. Another technical means can be an instance-based generation method. The computer system searches for instances similar to the current project in its stored historical tender document instance library (similarity can be judged based on multiple factors such as project type, scale, and technical requirements). For example, in a software project bidding, if the current project is to develop an enterprise resource management system, the computer system will search for tender document instances of similar enterprise resource management system development projects in the instance library. Then, the found instances are adjusted and modified according to the matching preset tender document generation framework to adapt to the specific requirements of the current project. Assuming that the project schedule portion of the found instance does not completely match the time requirements of the current project, the computer system will adjust that portion based on the duration requirements of the current project.

[0041] As an implementation mode, in step S100, before extracting implicit representations from a target text sequence set of a target bidding document and obtaining a first implicit representation set, the method may further include: Step S101: dividing the target bidding document into text sequences to obtain a first text sequence set and a second text sequence set, wherein the text sequences in the first text sequence set and the second text sequence set are arranged in reverse order to each other; Step S102: Walk the first text sequence set and the second text sequence set to obtain a target text sequence set.

[0042] Based on this, step S500, performs a bidding possibility analysis based on the second implicit representation set to obtain a target bidding support rate, including: step S500A: performs a bidding possibility analysis based on the second implicit representation set of the first text sequence set and the second implicit representation set of the second text sequence set to obtain a target bidding support rate.

[0043] In step S101, when the computer system processes the target bidding document, step S101 is first executed. The target bidding document is a document containing a lot of bidding-related information, which exists in the form of text. The computer system divides the text sequence into different viewpoints or sequences to organize the document content.

[0044] For example, the target bidding document contains: "The project requirements include the construction of a three-story building with modern office facilities. The construction must be completed within 6 months, and the quality must meet national building standards. In addition, the project budget is 5 million yuan." The computer system can divide it into text sequences according to certain rules. Assuming that the division is based on sentences, for the first text sequence set, the text sequence may be formed according to the original order of the sentences in the document, such as "The project requirements include the construction of a three-story building with modern office facilities." is one text sequence, "The construction must be completed within 6 months, and the quality must meet national building standards." is another text sequence, and "In addition, the project budget is 5 million yuan." is the third text sequence, etc.

[0045] The text sequences in the second text sequence set are arranged in reverse order with the text sequences in the first text sequence set. This means that the first text sequence in the second text sequence set is the reverse order of the last text sequence in the first text sequence set, and the second text sequence is the reverse order of the second to last text sequence in the first text sequence set, and so on. For the above example, in the second text sequence set, the last text sequence "In addition, the project budget is 5 million yuan." will become the first text sequence, but in reverse order, such as "500 yuan is the budget item, in addition" (here it is only for illustration of reverse order, and in actual processing, appropriate operations will be performed according to specific text processing technology), and so on.

[0046] Computer systems can use a variety of technical means to achieve this division. A simple method is to divide the sentences based on punctuation marks, and then construct two sets in order and reverse order respectively. If the text of the target bidding document is represented as T, the sentence boundaries can be determined by identifying punctuation marks such as periods and semicolons. (where t i Represents a single character), and when a punctuation mark is encountered, a sentence is divided into a text sequence.

[0047] In step S102, after obtaining the first text sequence set and the second text sequence set, the computer system executes step S102. Here, "walking" means that the computer system sequentially accesses each text sequence element in the set in a specific order.

[0048] For the first text sequence set, the computer system starts from the first text sequence and processes or analyzes them one by one until the last text sequence. For example, for each text sequence in the first text sequence set of the above construction project bidding, the computer system will sequentially process the text sequences about building construction requirements, construction period and quality requirements, project budget, etc. Similarly, for the second text sequence set, it is also processed sequentially starting from the first (first after reverse) text sequence.

[0049] During the walking process, the computer system can mark or number each text sequence for the convenience of subsequent operation and management. For example, the text sequences in the first text sequence set can be marked as (where m is the number of text sequences in the first text sequence set), and the text sequences in the second text sequence set are marked as (where n is the number of text sequences in the second text sequence set). Through this wandering operation, the computer system recombines or screens the text sequences in the two sets to obtain a target text sequence set. This target text sequence set contains the results of arranging the contents of the target tender documents from different arrangement order angles. For example, the target text sequence set may first contain part of the text sequences in the first text sequence set, and then contain part of the text sequences in the second text sequence set, and arrange them according to specific rules. This recombining method helps to analyze the contents of the target tender documents more comprehensively and deeply in subsequent steps (such as a series of operations starting from step S100).

[0050] In step S500A, when proceeding to step S500A, the computer system has obtained the second implicit representation sets corresponding to the first text sequence set and the second text sequence set respectively (this is the result of processing the two sets separately in the previous series of steps, such as steps S100-S400).

[0051] The second implicit representation set is a representation form of the original text sequence after complex feature extraction and transformation, which contains semantic information in the text sequence. For example, for each text sequence about the construction project in the first text sequence set (such as construction requirements, construction period requirements, etc.), after the previous processing, the corresponding elements in the second implicit representation set represent the relevant features of these text sequences in the form of vectors.

[0052] The computer system performs bidding possibility analysis based on these two second implicit representation sets. First, for the second implicit representation set of the first text sequence set, the computer system can analyze the factors related to bidding possibility reflected by each element (i.e., each second implicit representation). Taking a construction project as an example, if a second implicit representation shows that the technical capabilities of the bidder are highly matched with the project construction requirements (analyzed from the second implicit representation obtained by previously processing the construction requirements text sequence), this will have a positive impact on the bidding possibility.

[0053] Similarly, a similar analysis is performed on the second implicit representation set of the second text sequence set. For example, the second text sequence set may include a second implicit representation obtained by processing a text sequence related to the project budget. If this representation shows that the bidder's cost control ability can meet the project budget requirements, this is also a positive factor.

[0054] The computer system can perform bidding possibility analysis by constructing a comprehensive evaluation model. Suppose the second implicit representation set of the first text sequence set is (where p is the number of elements in the set), the second implicit representation set of the second text sequence set is (where q is the number of elements in the set). A function F can be defined to comprehensively evaluate the possibility of bidding, for example , where f is a function that analyzes a single second implicit representation to obtain a score associated with the likelihood of bidding, and is the weight given to the first text sequence set and the second text sequence set respectively, which is determined according to the actual situation of the project (for example, if the project focuses more on the technical requirements in the forward order text sequence, may be larger; if you are more concerned with factors such as budget in reverse order text sequences, may be larger).

[0055] The result calculated by this comprehensive evaluation model is the target bid support rate, which reflects the possibility of successful bidding after comprehensively considering the information contained in the two text sequence sets with different arrangement orders, and provides an important reference for whether to bid and how to generate the bidding documents (such as step S600) in the future.

[0056] As an implementation mode, in step S200, before iterating the yth feature accumulator according to the xth first implicit representation to obtain the xth feature accumulator, the method may further include: step S201: obtaining a discrete iteration variable corresponding to the continuous word embedding vector according to the lemma division step.

[0057] Based on this, in step S200, iterating the yth feature accumulator according to the xth first implicit representation to obtain the xth feature accumulator may include: step S210: fusing the xth first implicit representation and the yth feature accumulator according to discrete iteration variables to obtain the xth feature accumulator, and the fusion of the xth first implicit representation and the yth feature accumulator is used to complete the iteration of the yth feature accumulator.

[0058] In step S201, when the computer system executes step S201, it must first clarify the concept of lemma division step. Lemmatization is the process of dividing text into meaningful units (lemmas). For example, when processing the sentence "We are working on a construction project", lemma division may divide it into lemmas such as "we", "working", "a", and "construction project". The lemma division step is a measure of how many characters or lemmas are used to perform a specific operation during this division process.

[0059] Assume that the word quantization step is k. For a continuous sequence of word embedding vectors, the computer system processes it according to this step. Word embedding vector is a way of mapping word quanta to a low-dimensional vector space. For example, the word "architecture" may be mapped to an n-dimensional vector. . A continuous word embedding vector is a sequence of word embedding vectors corresponding to multiple word units in a text arranged in order.

[0060] Take "This building is very tall and has many rooms" as an example. After tokenization, we get tokens such as "this", "building", "tall", "there are", "many", and "rooms". Assume that each token has a corresponding 3D word embedding vector (just an example). If the tokenization step size k=2, then the computer system will consider the word embedding vectors corresponding to "this" and "building" as a group.

[0061] Discrete iterative variables are a type of variable obtained based on the above-mentioned lemma partitioning step and continuous word embedding vectors. It can represent a certain relationship or feature between continuous word embedding vectors at a specific step. The computer system can obtain discrete iterative variables through statistical operations or mathematical transformations. For example, for each group of word embedding vectors selected according to the lemma partitioning step, the sum of the Euclidean distances between them can be calculated as part of the discrete iterative variable. Assume and is the Euclidean distance between two consecutive word embedding vectors (selected according to the step size) , if there are m sets of such continuous word embedding vectors, then part of the discrete iteration variable D can be The computer system can also combine other statistical information, such as the mean and variance of the vector, to completely construct the discrete iterative variable.

[0062] When executing step S210, the computer system has obtained the discrete iteration variable, and will use it to fuse the xth first implicit representation and the yth feature accumulator to obtain the xth feature accumulator. The xth first implicit representation is part of the result obtained after the implicit representation extraction of the target bidding document in step S100. It contains the semantic information of a specific part of the text in the target bidding document after conversion. For example, when processing the bidding document for a construction project, if the xth first implicit representation corresponds to the text about the building structure requirements, then this implicit representation summarizes the information related to the building structure requirements in a vector form. The yth feature accumulator is an information storage structure that is gradually constructed and updated during the walk. It accumulates relevant information in the previous text sequence processing process. When the computer system reaches the xth first implicit representation, it will use the discrete iteration variable to fuse it with the yth feature accumulator.

[0063] Assume that the xth first implicit representation is , the y-th feature accumulator is , the discrete iteration variable is D. The computer system can use weighted summation to perform fusion. For example, first convert the discrete iteration variable D into a weight vector (Here k is a dimension determined by the specific composition of the discrete iteration variable.) Then, the new x-th feature accumulator It can be calculated by the following formula: .

[0064] This fusion method realizes the iteration of the yth feature accumulator, so that the xth feature accumulator contains both the new information in the current xth first implicit representation and the historical information in the previous yth feature accumulator. This helps to gradually integrate the information of different parts in the process of processing the target bidding document, laying the foundation for subsequent accurate bidding possibility analysis and bidding document generation. For example, in a construction project, this fusion can be used to integrate the information of the building structure requirements with the previously accumulated relevant information such as construction technology, project budget, etc., so that the feature accumulator can more comprehensively reflect the overall situation of the project.

[0065] As an implementation manner, step S500, performing bidding possibility analysis according to the second implicit representation set to obtain a target bidding support rate, may include: Step S510: Walk the second implicit representation set, and perform the following operations on the x-th second implicit representation walked to: Step S520: according to the xth second implicit representation, obtain the confidence of the xth target text sequence corresponding to the sth normal model to obtain the appropriate configuration confidence, where s is a positive integer less than or equal to S, and S is the number of normal models of the target bidding document; Step S530: re-representing according to the S adapted confidences of the x-th second implicit representation and the S first normal model coefficients of the S normal models to obtain an implicit representation of the to-be-analyzed text sequence; Step S540: Perform bidding possibility analysis based on the x implicit representations of the to-be-analyzed text sequences obtained by walking the second implicit representation set to obtain a target bidding support rate.

[0066] In step S510, the second implicit representation set is started to be walked. The second implicit representation set is obtained through a series of complex processing, and it contains the information representation of the target bidding document after specific transformation. For example, when processing a bidding for a construction project, each second implicit representation in this set may cover a comprehensive representation of different aspects of information such as project scale, technical requirements, budget, etc.

[0067] Walking means that the computer system visits each element in the second implicit representation set in a certain order, where each element is the xth second implicit representation (where x represents the sequence number of the current visit). This is similar to viewing each item in a list one by one, except that the items here are highly abstracted and transformed second implicit representations. The computer system can implement this walking operation through indexes or pointers, such as using loop structures in programming, starting from the first element of the collection and gradually moving to the next element until all elements are visited.

[0068] In step S520, for each x-th second implicit representation reached, the computer system executes step S520. The normal model here is a pre-built model for analyzing the target bidding document, and each normal model may correspond to different features or requirements in the target bidding document. For example, in the bidding of a construction project, there may be a normal model for analyzing the cost budget of the project, and another for analyzing the quality requirements of the project.

[0069] The computer system obtains the confidence level corresponding to the sth normal model according to the xth second implicit representation. The confidence level indicates the matching degree or conformity degree between the xth second implicit representation and the sth normal model. For example, assuming that the xth second implicit representation contains cost estimation information about a part of a construction project, and the sth normal model is a model about cost budget. The computer system can obtain the confidence level by calculating the distance or similarity between the two.

[0070] One possible technique is to calculate the Mahalanobis distance. Let the xth second implicit representation be the vector , the parameter of the sth normal model is the mean vector and the covariance matrix Then the Mahalanobis distance The calculation formula is The confidence level is then determined based on this distance. For example, the confidence level can be defined as for ,in is the preset maximum distance value.

[0071] Through such calculations, for each s-th normal model, the computer system can obtain a fitness confidence, which reflects the fitness of the x-th target text sequence (indirectly associated through the second implicit representation) with the s-th normal model.

[0072] In step S530 , the computer system performs re-representation using the S adapted confidences of the x-th second implicit representation obtained in step S520 and the first normal model coefficients of the S normal models.

[0073] The first normal model coefficients (such as mean variance) are important parameters of the normal model, which describe the distribution characteristics of the normal model. For example, for a normal model of cost budgeting, the mean may represent the expected average cost, and the variance represents the fluctuation range of the cost.

[0074] Assume that the S adapted confidences of the x-th second implicit representation are , the first normal model coefficients of the S normal models are (Here the mean and variance For example). Computer systems can perform re-representation by weighted summation.

[0075] Assume that the text sequence to be analyzed is implicitly represented as ,but ,here represents an operation to adjust the second implicit representation based on the normal model coefficients. For example, if the second implicit representation is in the form of a vector Operation can be .

[0076] Through this re-representation, the implicit representation of the text sequence to be analyzed fuses the information of different normal models and the adaptation information of the x-th second implicit representation and these normal models, thus providing a new representation for a more comprehensive and accurate analysis of the target bidding documents.

[0077] In step S540, when the computer system completes the walk of the second implicit representation set and obtains x implicit representations of the text sequence to be analyzed, in step S540, these representations are used to perform bidding possibility analysis to obtain the target bidding support rate.

[0078] Each implicit representation of the text sequence to be analyzed contains comprehensive information about different aspects of the target bidding document and the adaptation information to the normal model. The computer system can build a comprehensive evaluation model to perform bidding possibility analysis.

[0079] For example, suppose x text sequences to be analyzed are implicitly represented as The computer system can define a function F to calculate the bid likelihood score, , where f is a function that analyzes the implicit representation of a single text sequence to be analyzed and obtains a score, is the weight corresponding to the implicit representation of each text sequence to be analyzed, and this weight can be determined according to the importance of different parts of the target bidding document.

[0080] Then, this score is converted to the target bid support rate. For example, a mapping function P = g (F) can be used, where P is the target bid support rate and g is a function that maps the score to the 0-1 interval (indicating support rate). For example, ,in and are predetermined minimum and maximum score values.

[0081] Through this series of operations, the computer system starts from the second implicit representation set, and after steps such as walking, obtaining confidence, and re-representation, it finally obtains the target bid support rate. This ratio will provide an important reference for whether to bid in the future and how to generate the bidding documents.

[0082] This method of gradually deepening the analysis from the second implicit representation set makes full use of the converted information of the target bidding document, combines relevant knowledge such as the normal model, and comprehensively considers all aspects of the target bidding document, thereby improving the accuracy and reliability of the bidding possibility analysis.

[0083] As an implementation manner, step S520, obtaining the confidence of the xth target text sequence corresponding to the sth normal model according to the xth second implicit representation, and obtaining the adapted confidence may include: Step S521: obtaining S corrected confidences between the x-th second implicit representation and the S first normal model coefficients respectively; Step S522: fusing the S corrected confidences to obtain a corrected confidence fusion result; Step S523: among the S modified confidences, obtaining the target modified confidence corresponding to the first normal model coefficient of the s-th normal model and the x-th second implicit representation; Step S524: taking the weight of the target corrected confidence in the corrected confidence fusion result as the adapted confidence.

[0084] When the computer system executes step S521, the concept of the x-th second implicit representation and the first normal model coefficient is clarified. The x-th second implicit representation is an abstract representation of part of the target bidding document obtained through a series of previous processes, and it contains semantic information related to the target bidding document. For example, when processing a bidding document for a construction project, the x-th second implicit representation may cover a comprehensive vector representation of one or more aspects of information such as the construction project progress, quality requirements or cost budget.

[0085] The first normal model coefficient is a parameter used to describe the characteristics of the normal model. In this scenario, each normal model corresponds to different aspects or requirements in the target bidding document. Assume that in the construction project bidding, there is a normal model used to describe the cost budget, and its first normal model coefficient may include the mean (indicating the average cost budget) and the variance (indicating the fluctuation range of the cost budget); another normal model is used to describe the project quality requirements, and its first normal model coefficient may be some statistics related to the quality standard.

[0086] In order to obtain the S corrected confidences between the x-th second implicit representation and the S first normal model coefficients, the computer system needs to adopt specific technical means. One possible method is based on the calculation of the probability distribution function. Let the x-th second implicit representation be the vector , the first normal model coefficient of the sth normal model (s=1, 2, …, S) is (Based on the mean and variance for example).

[0087] For a univariate normal distribution, the probability density function is Each element in the x-th second implicit representation can be Treat it as an independent sample and calculate its probability density value under the sth normal model Then, the modified confidence between the xth second implicit representation and the sth normal model is obtained by some comprehensive method For example, we can use the weighted summation method and set the weight vector to ,but Through such calculations, the computer system can obtain S such modified confidences, each of which corresponds to the relationship between the x-th second implicit representation and a normal model.

[0088] In step S522, after obtaining S corrected confidences Afterwards, the computer system executes step S522 to fuse these modified confidences.

[0089] The purpose of this fusion is to synthesize the relationship between the x-th second implicit representation and all normal models into a single result for further analysis. The computer system can use a variety of fusion techniques, such as simple averaging, weighted averaging, or fusion methods based on other complex functions.

[0090] Taking the weighted average method as an example, let the weight vector be , then correct the confidence fusion result The formula can be Calculated. The weight here It can be determined based on the importance of different normal models in the entire target bidding document analysis. For example, in the construction project bidding, if the normal model of cost budget is relatively more important, then the weight corresponding to the modified confidence of the normal model of cost budget is (assuming s_0 is the serial number corresponding to the cost budget normal model) may be set relatively high.

[0091] In step S523, the computer system calculates the S corrected confidence scores obtained in step S523. , find the target corrected confidence associated with the sth normal model.

[0092] This target modified confidence is specifically for the sth normal model, and it reflects the specific relationship between the xth second implicit representation and the sth normal model. For example, in the example of construction project bidding, if the sth normal model is about the engineering quality requirements, then this target modified confidence represents a quantitative value of the degree of matching or correlation between the xth second implicit representation (which may contain some implicit information related to engineering quality) and the engineering quality requirements normal model. The computer system directly obtains the target modified confidence corresponding to the sth normal model from the S modified confidences through an index or matching mechanism. .

[0093] In step S524, the computer system corrects the confidence level of the target obtained in step S523. The modified confidence fusion result obtained in step S522 The weights in are determined as appropriate confidence.

[0094] The adaptation confidence indicates the degree of adaptation of the x-th target text sequence (indirectly associated through the x-th second implicit representation) to the s-th normal model. By calculating the weight of the target correction confidence in the fusion result, the computer system can more accurately measure this adaptation. For example, if the target correction confidence In the fusion results The larger the proportion of , the more important or more matching the relationship between the x-th second implicit representation and the s-th normal model is in the overall relationship, so the fit reliability is higher; on the contrary, if the smaller the proportion, the lower the fit reliability.

[0095] Assume the appropriate configuration confidence is ,but (Here we assume ). Through such calculation, the computer system obtains the appropriate confidence value, which will play an important role in the subsequent steps (such as the re-representation operation in step S530), and will help to further analyze the content of the target bidding document, thereby providing a more accurate basis for the final bidding possibility analysis and bidding document generation.

[0096] Through this series of steps S521-S524, the computer system gradually calculates the appropriate confidence level based on the relationship between the x-th second implicit representation and each normal model. This process makes full use of the characteristics of the normal model and the information contained in the second implicit representation, laying the foundation for a more comprehensive and accurate analysis of the target bidding documents.

[0097] As an implementation mode, in step S520, according to the xth second implicit representation, the confidence of the xth target text sequence corresponding to the sth normal model is obtained. After obtaining the adapted confidence, the method may further include: for the first normal model coefficient, loopingly performing the following processing: Step S520a: Obtain the second normal model coefficient of the sth normal model according to the x adapted confidences of the sth first normal model coefficient obtained by walking the second implicit representation set and the second implicit representation set; Step S520b: obtaining a target fitness confidence level based on the second normal model coefficient and the x-th second implicit representation iterative fitness confidence level; Step S520c: If the loop of iterative processing stops, then re-representation is performed based on the S target adaptation confidences of the x-th second implicit representation and the S second normal model coefficients of the S normal models to obtain an implicit representation of the text sequence to be analyzed.

[0098] The sth first normal model coefficients are predefined parameters used to describe the characteristics of the sth normal model, which have interacted with the xth second implicit representation in the previous step and generated a fit confidence. For example, in the context of a construction project bidding, if the sth normal model is about cost budgeting, the first normal model coefficients may include average cost (mean) and cost fluctuation range (variance). Fit confidence reflects the degree of fit between the xth second implicit representation and the sth normal model, and there is a corresponding fit confidence for each xth second implicit representation (as the second implicit representation set is walked).

[0099] The computer system uses the x adaptive confidences of the sth first normal model coefficients obtained by walking the second implicit representation set and the entire second implicit representation set to obtain the second normal model coefficients of the sth normal model. A possible technical approach is based on linear regression analysis.

[0100] Assume that the element in the second implicit representation set (i.e., the xth second implicit representation) is , the x appropriate confidences corresponding to the sth first normal model coefficient are The computer system can construct a linear regression model to fit the confidence is the dependent variable, expressed in the second implicit form Each element of is an independent variable.

[0101] For the linear regression model (in is the regression coefficient to be estimated, is the error term), the computer system estimates these regression coefficients by the least square method or other methods. Then, the second normal model coefficients of the sth normal model are constructed based on the estimated regression coefficients. For example, the second normal model coefficients can be some transformation related to the regression coefficients, such as (in yes in the second implicit representation set) and (This is just an example of how to build it, the actual situation may be more complicated).

[0102] Taking the bidding of construction projects as an example, if the elements in the second implicit representation set contain information about different aspects of the project, such as project scale, construction difficulty, etc., through the above linear regression analysis, the coefficients of the cost budget normal model (the sth normal model) can be adjusted according to the relationship between this information and the appropriate confidence level to obtain the coefficients of the second normal model that is more in line with the actual situation.

[0103] In step S520b, after obtaining the second normal model coefficients of the sth normal model, the computer system executes step S520b. The goal here is to iterate the previously obtained adaptive confidence using the newly obtained second normal model coefficients and the xth second implicit representation.

[0104] The purpose of iterative adaptation confidence is to update and optimize the original adaptation relationship based on the new model coefficient information. For example, continuing with the construction project bidding example, if the previously obtained adaptation confidence is calculated based on the initial normal model coefficient (first normal model coefficient), it reflects the degree of adaptation between the xth second implicit representation and the sth normal model (such as the cost budget model) under the initial model assumption. Now that there is a second normal model coefficient based on more information (such as the overall information of the second implicit representation set), this degree of adaptation needs to be re-evaluated.

[0105] The computer system can use a distance metric-based method to iteratively adapt the confidence. Let the x-th second implicit representation be , the second normal model coefficient of the sth normal model is The computer system can calculate some distance measure between the x-th second implicit representation and the new normal model (defined by the second normal model coefficients). For example, the Mahalanobis distance (in Is The vector composed of is with associated covariance matrix).

[0106] Then, the adaptation confidence is iterated based on this distance metric. For example, the target adaptation confidence can be defined as for (in is the preset maximum distance value). In this way, the target adaptation confidence comprehensively considers the new normal model coefficients and the information of the x-th second implicit representation, and more accurately reflects the adaptation relationship between them.

[0107] In step S520c, the loop may be stopped based on a preset stop condition, such as reaching a certain number of iterations, convergence of the objective function, etc.

[0108] At this point, the computer system has S target adaptation confidences for the xth second implicit representation And the S second normal model coefficients of the S normal models The computer system uses this information to perform a re-representation to obtain an implicit representation of the text sequence to be analyzed.

[0109] A possible re-representation method is based on weighted summation. Assume that the implicit representation of the text sequence to be analyzed is ,but .here represents an operation that adjusts the second implicit representation based on the normal model coefficients. For example, if , we can define .

[0110] Taking the construction project bidding as an example, if S normal models correspond to the project cost budget, engineering quality, construction progress and other aspects, then through this re-representation, the implicit representation of the text sequence to be analyzed is The x-th second implicit representation is integrated with the adaptation information of each normal model after iterative optimization and the new normal model coefficient information. This integrated representation can more comprehensively and accurately reflect the information related to the x-th second implicit representation in the target bidding document, providing a more valuable data basis for subsequent bidding possibility analysis and other operations.

[0111] As an implementation manner, step S520a, obtaining the second normal model coefficient of the sth normal model according to the x adapted confidences of the sth first normal model coefficient obtained by walking the second implicit representation set and the second implicit representation set, may include: Step S520a1: fusing the x adaptive confidences of the sth first normal model coefficient obtained by walking the second implicit representation set to obtain an adaptive confidence fusion result; Step S520a2: fusing the x adapted confidences with the second implicit representation set to obtain a confidence implicit representation fusion result; Step S520a3: Obtain a second normal model coefficient that has a change trend consistent with the confidence implicit representation fusion result and a change trend opposite to the adaptive confidence fusion result.

[0112] When executing step S520a1, the sth first normal model coefficient is determined when the normal model is previously constructed for the target bidding document, and is used to describe the parameters of certain characteristics of the normal model. The adaptation confidence is a value that measures the degree of adaptation between the xth second implicit representation and the sth normal model. Here, each xth second implicit representation (as the second implicit representation set is wandered) has a corresponding adaptation confidence, and there are x such adaptation confidences in total.

[0113] For example, in a bidding analysis of a large infrastructure construction project, assuming that the sth normal model is about the project schedule, the first normal model coefficients may include the expected average schedule duration, the acceptable fluctuation range of the schedule, etc. For each second implicit representation (the xth second implicit representation) related to the normal model of the project schedule obtained from the second implicit representation set walk, an adaptation confidence is calculated.

[0114] The purpose of the fusion operation of the computer system is to integrate the x adaptive confidences associated with the sth first normal model coefficient into a single result for subsequent operations. A variety of technical means can be used to achieve fusion, such as weighted averaging.

[0115] Assume that the x adaptive confidences are , the corresponding weight is (These weights can be determined based on the relative importance of different second implicit representations in the overall analysis. For example, if the target text sequence corresponding to a second implicit representation contains more critical item information, then its corresponding adaptation confidence weight can be higher.) Adaptation confidence fusion results It can be calculated by the following formula: ; In this way, the computer system combines x adaptive confidences into one value , this fusion result can comprehensively reflect the overall adaptation of all second implicit representations to the sth first normal model coefficient.

[0116] In step S520a2, the computer system merges the x adaptive confidences with the second implicit representation set. The second implicit representation set includes information representations of the target tender document after a series of processing, and each element (the xth second implicit representation) reflects part of the semantic content of the tender document to a certain extent.

[0117] Continuing with the example of infrastructure construction project bidding, assume that the second implicit representation of the elements in the set (x=1, 2, …), where each element may correspond to different aspects of the project, such as project scale, technical requirements, etc. It reflects the degree of fit of each second implicit representation to the sth normal model.

[0118] The computer system can adopt a fusion method based on vector multiplication and summation. First, each adaptive confidence The second implicit representation corresponding to Perform element-by-element multiplication to obtain a new vector Then, all these new vectors are summed up to get the confidence implicit representation fusion result . ; This fusion method incorporates the information of adaptive confidence into the second implicit representation set, making the fusion result It not only contains the semantic information in the original second implicit representation set, but also reflects the adaptation relationship information with the sth normal model. This helps to more comprehensively mine the features related to the target bidding document and the sth normal model, and provides a richer information basis for the subsequent acquisition of the second normal model coefficients.

[0119] In step S520a3, the computer system obtains the second normal model coefficient based on the existing fusion result. This involves implicitly expressing the fusion result with confidence. Adaptive confidence fusion results Analysis of changing trends.

[0120] The trend of change can be measured in many ways, such as calculating the first-order difference. For the confidence level, the fusion result is implicitly expressed , its first-order difference can be expressed as ; For the adaptive confidence fusion results , assuming that a series of Values, such as , then its first-order difference .

[0121] The second normal model coefficients that the computer system seeks must satisfy the conditions that the change trend of the confidence implicit representation fusion result is consistent with that of the confidence implicit representation fusion result and the change trend of the adaptive confidence fusion result is opposite. This means that if the confidence implicit representation fusion result shows an upward trend in a certain dimension (for example ), then the coefficients of the second normal model should also show changes in the corresponding dimensions that match this upward trend; if the adaptive confidence fusion result shows an upward trend at a certain stage (for example ), then the coefficients of the second normal model should show opposite trends in the corresponding aspects.

[0122] For example, in the bidding of infrastructure construction projects, if the confidence implicitly indicates that the dimension related to the project scale in the fusion result shows an increasing trend, this may mean that as the analysis of the bidding documents deepens, the relationship between the information related to the project scale and the sth normal model (such as the normal model of project progress) is strengthening. At this time, the part of the second normal model coefficient related to the project scale (for example, the coefficient of the impact of the project scale on the project progress) should match this growth trend. At the same time, if the adapted confidence fusion result shows that the part related to the project budget rises at a certain stage, and the change of the project budget may have a reverse impact on the project progress, then the part of the second normal model coefficient related to the project budget should show an opposite trend of change to reflect this complex relationship.

[0123] The computer system can determine the coefficients of the second normal model that meet this condition by establishing a mathematical model. For example, a model based on multivariate linear regression can be constructed, with the confidence implicit representation fusion result and the adaptive confidence fusion result as independent variables, and the second normal model coefficients as dependent variables, and the specific coefficient values ​​can be determined by analyzing the relationship in the data. is an n-dimensional vector, is a one-dimensional variable, and the coefficient of the second normal model is (m depends on the complexity of the normal model), then the model can be established (in is the parameter to be estimated, is the error term), and is estimated by the least squares method. , thereby obtaining the coefficients of the second normal model that meet the conditions.

[0124] Through steps S520a1-S520a3, the computer system gradually integrates and analyzes the information related to the sth first normal model coefficient, and finally obtains the second normal model coefficient. This process makes full use of the information in the adaptation confidence and the second implicit representation set, and provides the necessary basis for subsequent steps (such as the iteration of the adaptation confidence in step S520b, etc.), which helps to analyze the relationship between the target bidding document and the normal model more deeply and accurately, thereby improving the accuracy of the bidding possibility analysis.

[0125] As an implementation mode, before step S500, performing bidding possibility analysis based on the second implicit representation set to obtain the target bidding support rate, the method may also include: step S501: performing activation processing based on the first implicit representation set to obtain an implicit representation significant coefficient set.

[0126] Based on this, step S500, performing bidding possibility analysis according to the second implicit representation set to obtain a target bidding support rate, may include: Step S5001: merging the second implicit representation set and the implicit representation significant coefficient set to obtain an implicit representation set to be fused; Step S5002: Fusing the first implicit representation set and the implicit representation set to be fused to obtain a target implicit representation set; Step S5003: Perform bidding possibility analysis based on the target implicit representation set to obtain the target bidding support rate.

[0127] When executing step S501, the operation is first performed on the first implicit representation set. The first implicit representation set is the result of implicitly extracting the target text sequence set of the target bidding document in the previous step (such as step S100). Each element (first implicit representation) in this set summarizes the semantic information of the corresponding text sequence in the target bidding document in a vector form.

[0128] Activation processing is an operation that performs a nonlinear transformation on the elements in the first implicit representation set, with the purpose of highlighting some important features or information, thereby obtaining a set of implicit representation significant coefficients, which is essentially a set of weights used to represent the relative importance of each first implicit representation in subsequent analysis. For example, in a construction project bidding scenario, the target bidding document contains text sequences of different aspects such as the project scale, construction technical requirements, and construction period requirements. After implicit representation extraction, the elements in the first implicit representation set correspond to these different aspects of information vectors. The computer system can use activation functions (such as Sigmoid function, ReLU function, etc.) for activation processing.

[0129] Suppose a first implicit representation in the first implicit representation set is , if the Sigmoid function is used as the activation function, then the element j after activation is , the activated vector Each element in this activated vector is between 0 and 1, and is able to highlight certain elements with larger values ​​(indicating relatively more important information).

[0130] The computer system can be based on the activated vector Further calculation of the implicit representation of the significant coefficient set (weight set). For example, the vector The weight is determined by the proportion of each element in the entire vector. Let the implicit representation of the significant coefficient set be ,but In this way, the computer system obtains a set of implicit representation significance coefficients, each of which (weight) reflects the relative importance of the corresponding first implicit representation in the whole.

[0131] When executing step S5001, the second implicit representation set and the implicit representation significant coefficient set are already available. The second implicit representation set is another representation form of the target bidding document information obtained through a series of processing (such as steps S100-S400), which contains different processing results from the first implicit representation set, but also reflects various semantic information in the target bidding document.

[0132] The merging operation is to combine each element in the second implicit representation set with the corresponding weight in the implicit representation significant coefficient set. For example, let the second implicit representation set be , implicitly represents the set of significant coefficients as (Here we assume that m=n, the actual situation can be adjusted according to the specific processing method).

[0133] For each (j=1, 2, …, m), the computer system associates it with the corresponding weight Combine them to get a new element For example, if , ,So .

[0134] By performing such an operation on each element in the second implicit representation set, the computer system obtains the implicit representation set to be fused Each element in this set incorporates the weight information in the implicit representation significant coefficient set, so that each element can be reasonably integrated according to its importance in the subsequent fusion operation (step S5002).

[0135] In step S5002, the computer system fuses the first implicit representation set and the implicit representation set to be fused. The first implicit representation set contains the semantic information representation obtained by early processing of the target bidding document, while the implicit representation set to be fused is the result of fusing the weight information on the basis of the second implicit representation set.

[0136] The computer system can use a variety of methods to perform fusion, such as weighted summation. Let the first implicit representation set be , the implicit representation set to be fused is (Here we assume q=m, the actual situation can be adjusted according to the processing).

[0137] For each corresponding element and (k=1, 2, …, q), the computer system calculates the fused elements ,in and are the predetermined fusion weights (e.g. =0.4, = 0.6). For example, if , ,but .

[0138] By performing such a fusion operation on all corresponding elements, the computer system obtains the target implicit representation set This target implicit representation set integrates the information obtained from different processing stages of the target bidding document, including both the original semantic information in the early first implicit representation set and the information processed by the second implicit representation set and fused with weights, thus providing a richer information basis for a more comprehensive and accurate bidding possibility analysis.

[0139] In step S5003, the computer system performs a bidding possibility analysis based on the target implicit representation set obtained above to obtain the target bidding support rate. Each element in the target implicit representation set integrates various information of the target bidding document.

[0140] The computer system can build a model based on machine learning or statistical analysis to perform bidding possibility analysis. For example, a logistic regression model is used. Let the target implicit representation set be , these vectors are expanded into eigenvectors (r is the feature dimension after expansion).

[0141] The logistic regression model is of the form , where P is the target bid support rate, The computer system trains the logistic regression model through historical bidding data (known bidding document features and corresponding bidding success or failure marks) to determine the appropriate parameter values.

[0142] For example, in a construction project bidding, if some features in the target implicit representation set (such as the degree of match between the project scale and the bidder's capabilities, the degree of match between the construction technical requirements and the bidder's technical reserves, etc.) show a strong correlation with the success of the bid in historical data, then in the trained logistic regression model, the parameters corresponding to these features are This correlation will be reflected and thus given appropriate weight when calculating the target bid support rate.

[0143] Through such bidding possibility analysis, the target bidding support rate is obtained. This ratio can comprehensively reflect the feasibility of bidding based on the target bidding documents, and provide an important reference for subsequent bidding decisions (such as whether to generate bidding documents and how to generate bidding documents).

[0144] Through step S501 and a series of operations of steps S5001-S5003, the computer system starts from the first implicit representation set, undergoes activation processing, and multiple fusions with the second implicit representation set, and finally obtains the target bid support rate. This process makes full use of various information in the target bidding documents and improves the accuracy and reliability of the bid possibility analysis.

[0145] As an implementation mode, the target bid support rate is obtained based on the bid possibility analysis neural network processing, and the acquisition process of the bid possibility analysis neural network includes: Step S10: Perform bidding possibility analysis on the training text of the bidding document by initializing the neural network to obtain the predicted bidding support rate; Step S20: obtaining a target training cost according to an error between the predicted bid support rate and the bid support first verification mark of the training text of the bidding document; Step S30: Initialize the neural network according to the target training cost training to obtain the bidding possibility analysis neural network.

[0146] When executing step S10, the initialized neural network is a neural network model with initial parameter settings (such as initial weights and biases), which is designed to process tasks related to bidding possibility analysis.

[0147] For the training texts of the bidding documents, these texts contain various information related to the project bidding, such as project requirements, business terms, technical specifications, etc. The computer system inputs these training texts into the initialized neural network.

[0148] The neural network contains multiple neuron layers, such as input layer, hidden layer and output layer. The input layer receives the vector representation of the training text of the bidding document after some encoding (such as word vector encoding). The hidden layer processes the data by linearly combining the input data (weighted summation) and applying activation functions (such as ReLU function: f(x)=max(0, x)) to mine the complex relationships in the data.

[0149] Assume that the training text of the bidding document is encoded into a vector , the weight matrix from the input layer to the hidden layer is , the bias vector of the hidden layer is , then the output of the hidden layer The formula can be After being processed by multiple hidden layers, the predicted bid support rate is finally obtained in the output layer. For example, if the output layer has only one neuron, then the output value y of this neuron is the predicted bid support rate, which represents the prediction of the possibility of successful bidding for the training text of the bidding document based on the current initialized neural network. The value of y is between 0 and 1, and the closer it is to 1, the higher the possibility of successful bidding.

[0150] Take a construction project bidding as an example. The training text contains information such as project scale, construction period requirements, budget, etc. The neural network analyzes and processes this information according to its initial parameter settings to obtain a predicted bid support rate.

[0151] In step S20, the computer system compares the difference between the predicted bid support rate and the priori mark. The priori mark is a known actual bid support rate corresponding to the training text of the bidding document or a mark indicating whether the bid is successful or not (for example, a successful bid is marked as 1 and a failed bid is marked as 0).

[0152] There are many methods for error calculation, such as mean square error (MSE). Assume that the predicted bid support rate is , the prior is marked as y, then the mean square error (where n is the number of training samples.) This mean square error is a representation of the target training cost.

[0153] In step S30, the computer system uses the target training cost obtained in step S20 to adjust the parameters of the initialized neural network, thereby obtaining a bidding possibility analysis neural network.

[0154] A common training method is to use the gradient descent algorithm, which takes the derivative of the parameters (weights and biases) in the neural network based on the target training cost (such as mean squared error) to find the parameter values ​​that minimize the target training cost.

[0155] Assume that the parameters of the neural network are (including all weights and biases), the target training cost is According to the gradient descent algorithm, the parameter update formula is ,in is the learning rate, which determines the step size of each parameter update. is the target training cost with respect to the parameter gradient.

[0156] For example, for an element in the weight matrix W , calculate its gradient , and then according to the learning rate To update By performing such update operations on all parameters in the neural network multiple times, the target training cost is continuously reduced until the preset stopping condition is reached (such as the target training cost is less than a certain threshold, or the maximum number of training rounds is reached).

[0157] After this training process, the parameters of the initialized neural network are optimized, thus becoming a bidding possibility analysis neural network that can perform bidding possibility analysis more accurately. When this neural network subsequently performs bidding possibility analysis on new target bidding documents, it can more reasonably analyze various information and output accurate bidding support rates based on the parameter patterns obtained from previous training.

[0158] As an implementation mode, step S30, training and initializing a neural network according to a target training cost to obtain a bidding possibility analysis neural network, may include: Step S31: obtaining a third normal model coefficient set iterated from each of the R tender document training texts, obtaining R third normal model coefficient sets, the third normal model coefficient set including S third normal model coefficients corresponding to the S normal models, R≥2; Step S32: fusing R third normal model coefficient sets to determine a normal model coefficient fusion set; Step S33: combining the initial normal model coefficient set and the normal model coefficient fusion set of the bidding possibility analysis neural network to obtain a normal model coefficient iteration set; Step S34: adjusting the network internal configuration variables of the initialized neural network according to the target training cost to obtain updated network internal configuration variables; Step S35: Iterate the normal model coefficients and update the internal configuration variables of the network to obtain a neural network for bidding possibility analysis.

[0159] When executing step S31, the tender document training text is the basic data used to train the bidding possibility analysis neural network, and there are R such training texts (R ≥ 2). The normal model is a mathematical model that is used to describe various characteristics or indicators related to the tender in this context. Each normal model has a corresponding coefficient, which is called the third normal model coefficient here.

[0160] For each tender document training text, the computer system will iterate to obtain the third normal model coefficient set. For example, suppose in the scenario of construction project bidding, there are 3 (R=3) tender document training texts. One is about a large commercial construction project, another is about a small residential construction project, and another is about an industrial construction project.

[0161] For these projects, it is assumed that there are 2 (S=2) normal models, one for cost budget and the other for project schedule. For the training text of the bidding documents for large commercial construction projects, during the processing, a set of third normal model coefficients may be iterated for the cost budget normal model, such as the estimated value of the average cost, the fluctuation range of the cost, etc.; for the project schedule normal model, coefficients such as the average schedule duration and the acceptable fluctuation range of the schedule will also be iterated. Similarly, similar operations are performed on the training text of the bidding documents for small residential construction projects and industrial construction projects, and finally 3 (R) sets of third normal model coefficients are obtained, each set containing the third normal model coefficients corresponding to 2 (S) normal models.

[0162] The computer system can use statistical analysis and data mining techniques to perform this iterative operation. For example, for the cost budget normal model, the third normal model coefficient can be determined by calculating statistics such as mean and variance based on the cost estimates and cost components of different project parts in the training text. For the project schedule normal model, the relevant coefficient can be determined by analyzing the duration requirements, milestone plans and other information in the text.

[0163] In step S32, after obtaining R third normal model coefficient sets, the computer system executes step S32 to perform a fusion operation. The purpose of fusion is to integrate the information in all the training texts of the bidding documents to obtain a more comprehensive and representative normal model coefficient set, that is, a normal model coefficient fusion set.

[0164] Continuing with the above construction project bidding example, for the cost budget normal model, assuming that from the third normal model coefficient set of 3 (R=3) different projects (large commercial buildings, small residential buildings, industrial buildings), each set has a coefficient on the average cost estimate. The computer system can use the weighted average method to merge.

[0165] Suppose the coefficient of the third normal model coefficient set for the first project (large commercial building) on ​​the average cost estimate is , whose weight is (The weights may be determined based on the size, complexity, or other relevant factors of the project; for example, a large commercial building project may be given a higher weight because of its size and complexity). The corresponding coefficient for the second project (a small residential building) is , the weight is ; The third project (industrial building) is , the weight is .

[0166] Then, in the normal model coefficient fusion set, the fusion coefficient of the average cost estimate of the cost budget normal model is The formula can be Calculated.

[0167] For the normal model of project progress and other possible normal models (S), they are calculated according to a similar weighted average method to obtain the fusion set of the entire normal model coefficients. This fusion method can comprehensively consider the characteristics of different project types, so that the fused coefficients can better reflect the normal model characteristics under general circumstances.

[0168] In step S33, the initial normal model coefficient set of the bidding possibility analysis neural network is combined with the normal model coefficient fusion set obtained previously. The bidding possibility analysis neural network has a set of initial normal model coefficient sets when initialized. These coefficients are set in the network initialization stage and may be based on some prior knowledge or simple estimation.

[0169] For example, when the neural network is initialized, the initial average cost estimate coefficient set for the cost budget normal model is , the initial average progress duration coefficient set for the normal model of project progress is Etc. (assuming there are S initial coefficients corresponding to the normal model).

[0170] One way to combine the initial normal model coefficient set and the normal model coefficient fusion set can be weighted summation. Let the fusion coefficient be , for the cost budget normal model, the corresponding coefficient in the normal model coefficient iteration set The formula can be Calculated, where It is the coefficient of the cost budget normal model in the normal model coefficient fusion set obtained in step S32.

[0171] Similarly, the coefficients corresponding to the other S-1 normal models are calculated in a similar way to obtain the normal model coefficient iteration set. This combination not only takes into account the initial settings of the neural network, but also integrates the information mined from the training text of the bidding documents, so that the coefficient iteration set can be optimized and adjusted on the original basis.

[0172] In step S34, the internal configuration variables of the initialized neural network are adjusted according to the target training cost. The target training cost is calculated in step S20 according to the error between the predicted bid support rate and the priori mark, which reflects the degree of deviation between the predicted result of the current neural network and the actual result.

[0173] The internal configuration variables of the network mainly include the connection weights and bias terms between neurons. For example, for a simple neural network structure, assuming that the input layer has n neurons, the hidden layer has m neurons, the weight matrix connecting the input layer and the hidden layer is W (W is an m×n matrix), and the bias vector of the hidden layer is (m-dimensional vector).

[0174] The computer system can use the gradient descent algorithm to adjust these variables. For example, we first need to calculate the target training cost J about The partial derivative of The calculation of this partial derivative involves the forward propagation (calculating the prediction result from the input layer to the output layer) and back propagation (calculating the impact of the error on the parameters of each layer from the output layer based on the error between the prediction result and the target) algorithms of the neural network.

[0175] Assume that the learning rate is (a small positive number that is pre-set to control the step size of the adjustment), then The update formula is .

[0176] For the bias vector Elements in , and its partial derivatives are calculated in a similar way , and then by to update.

[0177] By performing such an update operation on all the connection weights and bias items within the network, the computer system obtains and updates the internal configuration variables of the network. This process is repeated continuously, so that the prediction ability of the network is gradually improved, with the goal of minimizing the target training cost.

[0178] In step S35, the computer system combines the iterative set of normal model coefficients with the updated network internal configuration variables to obtain the final bidding possibility analysis neural network.

[0179] The normal model coefficient iteration set includes coefficients related to S normal models, which reflect information about the characteristics of the bidding project (such as cost, schedule, etc.) obtained from the training text of the bidding document. The updated network internal configuration variables are the parameters such as weights and biases of the neural network adjusted according to the target training cost in step S34.

[0180] For example, in the final bidding possibility analysis neural network, for the part that processes cost budget related information, the coefficients of the cost budget normal model in the normal model coefficient iteration set and the updated weights and bias parameters will be used to build the corresponding calculation logic. Similarly, for other aspects of information processing such as project progress, the corresponding normal model coefficients and updated network parameters will also be comprehensively considered.

[0181] This joint operation enables the bidding possibility analysis neural network to have both the project characteristics related information mined from a large number of bidding document training texts (through the iterative set of normal model coefficients) and an optimized and adjusted network structure (by updating the internal configuration variables of the network), so that it can more accurately perform bidding possibility analysis on new bidding documents.

[0182] As an implementation mode, step S600, based on the target bid support rate, calling a matching preset bid document generation framework to generate a bid document may include: Step S610: calling the target bidding document generation network trained in advance; Step S620: input the general bidding text and the matching preset bidding document generation framework into the target bidding document generation network, and obtain the bidding document based on the target bidding document generation network.

[0183] In step S610, it is necessary to call the target bidding document generation network that has been trained in advance. This target bidding document generation network is a neural network structure specifically used to generate bidding documents based on input information. In the previous training process (related to steps S1-S4), it has learned how to generate appropriate bidding document content based on different inputs (such as bidding training text, sample bidding document generation framework, etc.).

[0184] For example, in the field of construction project bidding, this network may have learned how to generate corresponding bidding document content based on information such as the type of construction project (residential, commercial, industrial, etc.), scale, technical requirements, etc. This network contains multiple layers of neurons, and the neurons are connected by different weights. These weights are determined during the training process and can reflect the relationship between different input features and the content of the bidding document.

[0185] In step S620, the computer system provides the general bidding text and the matching preset bidding document generation framework as input to the target bidding document generation network. The general bidding text contains some basic general information about the bidding, such as the basic situation of the bidder (company size, qualifications, etc.), basic understanding of the project, etc.

[0186] A preset bid document generation framework is a predefined structure or template that specifies what parts a bid document should contain and the general structure of these parts. For example, a typical preset bid document generation framework may include project overview, solution, project schedule, quotation strategy, etc.

[0187] After the computer system passes these input information to the target bid document generation network, the neurons in the network process these inputs according to the pre-learned weights. For example, for the information about the bidder's qualifications in the input general bid text, the neurons in the network will judge its importance in generating different parts of the bid document content according to the weights. If the qualification information is more important when generating the project overview, then the neuron connection weights related to the qualification will make this part of the information more prominent in the project overview.

[0188] The network processes the bids layer by layer, gradually generating the various parts of the bid document, and finally combining them into a complete bid document. This process may involve complex calculations. For example, when generating the quotation strategy, it may be necessary to generate a reasonable quotation based on the project scale, technical requirements and other information in the input information, combined with the cost calculation model learned by the network.

[0189] Among them, the training process of the target bidding document generation network includes: Step S1: Obtain bidding training text and sample bidding document generation framework; Step S2: in the bidding document generation network to be trained, greedily search the bidding training text through the example bidding document generation framework to obtain a reduced dimension representation vector of the training text; Step S3: In the bidding document generation network to be trained, a beam search is performed on the reduced dimension representation vector of the training text through the example bidding document generation framework to obtain the increased dimension representation vector of the training text, and a weighted fusion is performed on the reduced dimension representation vector of the training text and the increased dimension representation vector of the training text through the initial fusion variable to obtain a training prediction representation vector; Step S4: Generate a cost function based on the training prediction representation vector and the bidding training text, optimize the initial fusion variable based on the cost function until the representation vector fusion variable is obtained when the variable is stable, and determine the bidding document generation network to be trained including the representation vector fusion variable as the target bidding document generation network; the target bidding document generation network is used to generate a bidding document based on the bidding document generation framework.

[0190] In step S2, the computer system operates the bidding document generation network to be trained. The guiding parameters in the example bidding document generation framework are preset parameters for guiding the search process. For example, these parameters may specify the priority order when searching for important information in the bidding training text. For example, greedy search is a search strategy that selects the option that currently appears to be the best in each selection step. The computer system performs a greedy search on the bidding training text through the guiding parameters in the example bidding document generation framework. For example, during the search process, if the guiding parameters set the core technical requirements of the project as a high priority, the computer system will first look for relevant information about the core technology of high-rise residential buildings (such as structural system, seismic design, etc.) in the bidding training text.

[0191] During the search process, the computer system gradually converts the information in the bidding training text into a reduced-dimensional representation vector. Assuming that the information in the bidding training text is high-dimensional (containing many words, sentences and other information elements), through the greedy search process, the computer system compresses and refines this information according to certain rules and weights to obtain a low-dimensional training text reduced-dimensional representation vector. This vector can represent the key information in the bidding training text in a more concise form, which is convenient for subsequent network processing.

[0192] In step S3, the computer system first uses a beam search method. Beam search is an improved search strategy that retains multiple (number of beam widths) optimal choices at each search step, rather than selecting only one optimal choice like greedy search. Through the example bidding document generation framework, a beam search is performed on the dimensionality reduction representation vector of the training text in the training bidding document generation network. For example, in the scenario of generating a high-rise residential building bidding document, the beam search may consider multiple candidate solutions for the project solution (rather than just selecting a solution that appears to be the best), and these candidate solutions all meet the requirements of the example bidding document generation framework to a certain extent.

[0193] In this process, the computer system expands the reduced-dimensional representation vector of the training text into an increased-dimensional representation vector of the training text based on the searched multiple candidate solutions and the structure and parameter requirements in the sample bidding document generation framework. This increased-dimensional representation vector contains more detailed information about the content of different parts of the bidding document. It is an enrichment and expansion of the reduced-dimensional representation vector based on considering multiple candidate solutions.

[0194] Then, the computer system uses the initial fusion variable to perform weighted fusion on the training text dimension reduction representation vector and the training text dimension increase representation vector. The initial fusion variable is a pre-set weight parameter used to control the relative importance of the two vectors in the fusion process. Suppose the training text dimension reduction representation vector is , the training text dimension-raising representation vector is , the initial fusion variable is (0 < < 1), then the training prediction representation vector The formula can be This training prediction representation vector combines the information of the dimension reduction and dimension increase representation vectors, and is a preliminary prediction representation of the content of the bidding document by the bidding document generation network to be trained.

[0195] In step S4, the cost function is used to measure the difference between the training prediction representation vector and the actual bidding document content. For example, the mean square error (MSE) can be used as the cost function, and the representation vector corresponding to the actual bidding document is ,but (where n is the dimension of the vector).

[0196] The computer system then optimizes the initial fusion variable based on this cost function. The optimization process is usually iterative, for example using a gradient descent algorithm. In each iteration, the value of the initial fusion variable is adjusted based on the partial derivative of the cost function with respect to the initial fusion variable, so that the value of the cost function gradually decreases. This process is repeated until the initial fusion variable reaches a stable state, that is, further adjustments to it will not significantly improve the cost function.

[0197] When the representation vector fusion variable is obtained when the variable is stable, the bidding document generation network to be trained containing this stable representation vector fusion variable is determined as the target bidding document generation network. This target bidding document generation network has been optimized and trained, and can generate relatively accurate bidding document content according to the input bidding document generation framework. For example, when facing a new high-rise residential building bidding project, it can generate a bidding document that meets the requirements based on the input relevant information, including reasonable project overview, solutions, quotation strategy and other parts.

[0198] As an implementation method, step S1, obtaining a bidding training text and a sample bidding document generation framework, may include: Step S11: obtaining a first training target, and obtaining a second training target corresponding to the first training target; the second training target includes the first training target; Step S12: obtaining a first bidding document generation network corresponding to the second training target, using the first bidding document generation network as a basic bidding document generation network for the first training target, and obtaining a training sample library corresponding to the first training target and the basic bidding document generation network; the basic bidding document generation network is a trained bidding document generation network; Step S13: obtaining bidding training texts having features corresponding to the first training target in the training sample library, and generating a sample bidding document generation framework based on the first training target.

[0199] In step S11, the computer system first determines the first training goal. The first training goal is a relatively specific and targeted goal, which clarifies a specific direction of the bidding document generation network training. For example, in the field of construction project bidding, the first training goal may be "generating bidding documents for large commercial complex construction projects." The "large commercial complex construction project" here is a specific type, which has unique characteristics such as building scale, functional requirements, and business operation model.

[0200] Then, the computer system obtains a second training goal corresponding to the first training goal. The second training goal is a broader concept that includes the first training goal. Continuing with the construction project as an example, the second training goal may be "generate bidding documents for all commercial construction projects." Commercial construction projects include large commercial complexes, shopping malls, office buildings and other types, so this second training goal covers the large commercial complex construction projects in the first training goal.

[0201] The computer system can obtain these two targets through a pre-established project type hierarchy or classification system. For example, in a construction project classification database, different types of construction projects are classified and stored in order from specific to broad. The computer system can query this database and find the corresponding broader category (such as "commercial construction project") as the second training target based on the keyword of the first training target (such as "large commercial complex construction project").

[0202] In step S12, firstly, the first bid document generation network corresponding to the second training target is found. This first bid document generation network is obtained after a certain training, and is specifically used to handle the bid document generation task related to the second training target. For example, in the context of construction project bid document generation, if the second training target is "generate bid documents for all commercial construction projects", then the corresponding first bid document generation network has learned some general rules, patterns and features of commercial construction project bid document generation.

[0203] The computer system uses the first bidding document generation network found as the base bidding document generation network for the first training goal (e.g., "generate bidding documents for a large commercial complex construction project"). This is because although the first training goal is more specific, it is part of the second training goal, so this already trained, broader network can be used as a basis for further training.

[0204] Next, the computer system obtains a training sample library related to the first training target and the basic bidding document generation network. This training sample library contains a large amount of sample data related to the first training target, which is essential for training the bidding document generation network. Taking a large commercial complex construction project as an example, the training sample library may contain many past large commercial complex construction projects' bidding documents, related project requirements documents, bidder's solution descriptions and other materials. The computer system can obtain this data through the indexing and search functions in the data storage system. For example, if all bidding-related data are stored in a large database, the computer system can retrieve the corresponding training sample library from the database based on the keywords of the first training target (such as "large commercial complex construction project") and the identification information related to the first bidding document generation network.

[0205] When the computer system executes step S13, it first selects the bidding training texts with the features corresponding to the first training target from the previously acquired training sample library. In the example of a large commercial complex construction project, these features may include an extremely large building area, complex functional partitions (such as the integration of multiple functions such as business, office, and entertainment), and high-standard building facilities (such as large parking lots, high-end fire protection and security systems, etc.). The computer system analyzes each sample in the training sample library to determine whether it contains these specific features, thereby selecting bidding training texts that meet the requirements. For example, this screening process can be achieved through feature extraction and matching technology. Assuming that each sample is stored in text form, the computer system can define a set of keywords or phrases related to the features of large commercial complex construction projects (such as "building area exceeds [X] square meters", "contains commercial, office and entertainment functional partitions", etc.), and then perform keyword matching on each sample text, count the number of matches or the importance of the matches, and when a certain threshold is reached, the sample is determined as a bidding training text with the features corresponding to the first training target.

[0206] Based on the selected bidding training texts, the computer system begins to build a sample bidding document generation framework. This framework is generated according to the specific requirements of the first training goal. It specifies which parts the bidding document should contain and the general structure and content focus of these parts. For the bidding document generation framework for large-scale commercial complex construction projects, the project overview section may emphasize the scale of the project and the complexity of the functional combination; the solution section may elaborate on how to meet multiple functional requirements while ensuring the overall coordination and safety of the building; and the quotation strategy section may consider the cost structure characteristics of large projects (such as land costs, large-scale construction costs, equipment and facility procurement costs, etc.). The computer system can build this framework by analyzing the content structure and common patterns in the bidding training text. For example, the frequency of occurrence of different parts of the content, the logical relationship between the content, etc., and then determine the structure and key content of the framework based on these statistical information. At the same time, some industry standards and best practices can be referred to to improve this sample bidding document generation framework to ensure its rationality and effectiveness.

[0207] As an implementation method, the method may further include: Step S14: if the first bidding document generation network corresponding to the second training target is not obtained, the basic network composed of the networks corresponding to the first training target is used as the basic bidding document generation network, and the initial fusion variables are added to the basic bidding document generation network; Step S15: If the first bidding document generation network corresponding to the second training target is obtained, a step of using the first bidding document generation network as a basic bidding document generation network for the first training target is performed.

[0208] If the computer system does not obtain the operation of the first bidding document generation network corresponding to the second training target. The first bidding document generation network here refers to a network that has been trained and can be used for a specific bidding document generation task. For example, assuming that the first training target is to generate bidding documents for "large-scale construction engineering projects" (here "large-scale construction engineering projects" is the first training target), and the second training target is to generate bidding documents for "all construction engineering projects" ("all construction engineering projects" include "large-scale construction engineering projects", which is the inclusion relationship between the second training target and the first training target). If the computer system does not obtain the bidding document generation network (the first bidding document generation network) that has been trained for "all construction engineering projects" (the second training target), then the computer system will use the basic network composed of the network corresponding to the first training target as the basic bidding document generation network. This basic network may be constructed by some basic modules related to the first training target, such as a feature extraction module related to the basic elements of construction engineering projects. At the same time, the computer system will add initial fusion variables to this basic bidding document generation network. This initial fusion variable may be a parameter used for weighted fusion of different representation vectors in subsequent steps, for example, it can be expressed as α, and its initial value can be set to a certain value based on experience, such as α=0.5.

[0209] If the computer system obtains the operation of the first bidding document generation network corresponding to the second training target. Continuing with the above example, if the computer system has obtained the trained bidding document generation network (first bidding document generation network) for "all construction engineering projects" (the second training target), then the computer system will use this first bidding document generation network as the basic bidding document generation network for the first training target ("large-scale construction engineering projects"). The advantage of doing so is that, since "all construction engineering projects" include "large-scale construction engineering projects", the bidding document generation network for a wider range of project types may already include some common features and functions applicable to specific types (large-scale projects), thereby better providing a basis for generating bidding documents of a specific type ("large-scale construction engineering projects").

[0210] When the computer system determines whether to obtain the first bidding document generation network corresponding to the second training target, it can be achieved by querying a pre-established network repository. This repository stores information such as different types of bidding document generation networks and their corresponding target types. When constructing a basic network as the basic bidding document generation network, the computer system can combine corresponding basic modules according to the characteristic definition of the first training target. For example, if the first training target is "large-scale construction engineering projects", then the basic modules may include cost estimation modules for large projects, schedule scheduling modules for large projects, etc. These modules can be combined into a basic network through specific interfaces and connections. As for the addition of initial fusion variables, it can be achieved by adding an adjustable parameter node in the network structure definition. The value of this parameter node (such as ) will be adjusted according to factors such as the cost function in the subsequent optimization process.

[0211] As an implementation mode, step S4, optimizing the initial fusion variables based on the cost function until obtaining the representation vector fusion variables when the variables are stable, may include: Step S41: if the basic bidding document generation network is the first bidding document generation network, the initial fusion variables are optimized based on the cost function until a representation vector fusion variable is obtained when the variable is stable; Step S42: If the basic bidding document generation network is not the first bidding document generation network, then based on the cost function, the initial configuration variables and the initial fusion variables in the basic bidding document generation network are optimized until the representation vector fusion variables and the network configuration variables are obtained when the variables are stable; the initial configuration variables are the configuration variables in the basic bidding document generation network, the network configuration variables are the variables after the initial configuration variables are stabilized, and the representation vector fusion variables are the variables after the initial fusion variables are stabilized.

[0212] For step S41, when the computer system determines that the basic bidding document generation network is the first bidding document generation network, the computer system optimizes the initial fusion variables based on the cost function until the representation vector fusion variables are obtained when the variables are stable. The basic bidding document generation network here is the first bidding document generation network, which means that there is a network that has been pre-trained for a specific training goal (such as the goal of generating a certain type of bidding document mentioned earlier). For example, assuming that the first training goal is to generate a bidding document for an "electronic equipment procurement project", the first bidding document generation network is a network specially trained for generating such bidding documents. The initial fusion variable is a parameter used to perform weighted fusion of different representation vectors (such as the training text dimensionality reduction representation vector and the training text dimensionality increase representation vector) in the bidding document generation network, and is set to The cost function is a function used to measure the difference between the training prediction representation vector and the bidding training text, which can be expressed as The computer system is adjusted The value of the cost function is minimized A common technique is to use the gradient descent method, which is ,in is current value, It is the next iteration value, is the learning rate. The computer system is constantly iterating , until the cost function Convergence, at this time It is the representation vector fusion variable when the variable is stable.

[0213] For step S42, when the computer system determines that the basic bidding document generation network is not the first bidding document generation network, the computer system optimizes the initial configuration variables and initial fusion variables in the basic bidding document generation network based on the cost function until the representation vector fusion variables and network configuration variables are obtained when the variables are stable. Assume that the basic bidding document generation network is a general network framework (not a network specifically for the current first training target), for example, a basic bidding document generation framework applicable to various types of projects. The initial configuration variables are inherent parameters in this basic network framework, such as the number of layers in the network, the number of neurons in each layer, etc., and the initial configuration variables are set to (here can be a vector representing multiple configuration parameters), the initial fusion variable is still set to Cost function The influence of the initial fusion variables and the initial configuration variables on the difference between the training prediction representation vector and the bidding training text is also considered. The computer system can also use a method similar to gradient descent, but this time it is to simultaneously and Make adjustments. , the formula is similar to the above gradient descent formula ;for , ,in and They are respectively for and The learning rate of the computer system is constantly iterating and , until the cost function Convergence, at this time It is the representation vector fusion variable when the variable is stable. That is, the network configuration variables. In this way, by processing different situations, the computer system can reasonably optimize the relevant variables according to the situation of the basic bidding document generation network, thereby constructing a target bidding document generation network that is more suitable for the bidding document generation task.

[0214] The present application embodiment provides a computer system, such as Figure 2 As shown, the computer system 100 includes: a processor 101 and a memory 103. The processor 101 and the memory 103 are connected, such as through a bus 102. Optionally, the computer system 100 may also include a transceiver 104. It should be noted that in actual applications, the transceiver 104 is not limited to one, and the structure of the computer system 100 does not constitute a limitation on the embodiments of the present application.

[0215] An embodiment of the present application provides a computer system. The computer system in the embodiment of the present application includes: one or more processors; a memory; and one or more computer programs, wherein the one or more computer programs are stored in the memory and configured to be executed by the one or more processors, and when the one or more programs are executed by the processor, the above method is implemented.

Claims

1. A method for intelligently generating bidding documents based on natural language processing, characterized in that: The method includes: performing implicit representation extraction on a target text sequence set of a target bidding document to obtain a first implicit representation set; walking the first implicit representation set, and performing the following operations on the xth first implicit representation walked to, wherein x is a positive integer less than X, and X is the number of target text sequences in the target text sequence set: iterating the yth feature accumulator according to the xth first implicit representation to obtain the xth feature accumulator, if x is 1, the yth feature accumulator is the initial feature accumulator, if x is greater than 1, the yth feature accumulator is the feature accumulator of the yth target text sequence, wherein y=x-1; obtaining the xth second implicit representation according to the xth feature accumulator; obtaining the second implicit representation set according to the x second implicit representations obtained by walking the first implicit representation set; performing bidding possibility analysis according to the second implicit representation set to obtain a target bidding support rate; based on the target bidding support rate, calling a matching preset bidding document generation framework to generate a bidding document.

2. The method according to claim 1, characterized in that Before performing implicit representation extraction on the target text sequence set of the target bidding document to obtain the first implicit representation set, the method also includes: performing text sequence division on the target bidding document to obtain a first text sequence set and a second text sequence set, wherein the text sequences in the first text sequence set and the second text sequence set are arranged in reverse order to each other; performing walks on the first text sequence set and the second text sequence set to obtain the target text sequence set; performing bidding possibility analysis based on the second implicit representation set to obtain the target bidding support rate includes: performing bidding possibility analysis based on the second implicit representation set of the first text sequence set and the second implicit representation set of the second text sequence set to obtain the target bidding support rate.

3. The method according to claim 1, characterized in that Before iterating the yth feature accumulator according to the xth first implicit representation to obtain the xth feature accumulator, the method further includes: obtaining a discrete iteration variable corresponding to a continuous word embedding vector according to a lemma division step; iterating the yth feature accumulator according to the xth first implicit representation to obtain the xth feature accumulator, including: fusing the xth first implicit representation and the yth feature accumulator according to the discrete iteration variable to obtain the xth feature accumulator, and the fusion of the xth first implicit representation and the yth feature accumulator is used to complete the iteration of the yth feature accumulator; performing a bidding possibility analysis according to the second implicit representation set to obtain a target bidding support rate, The method comprises: walking the second implicit representation set, and performing the following operations on the x-th second implicit representation walked to: obtaining the confidence of the x-th target text sequence corresponding to the s-th normal model according to the x-th second implicit representation, and obtaining the adaptive configuration confidence, wherein s is a positive integer less than or equal to S, and S is the number of the normal models of the target bidding document; re-representing according to the S adaptive configuration confidences of the x-th second implicit representation and the S first normal model coefficients of the S normal models, and obtaining the implicit representation of the text sequence to be analyzed; performing bidding possibility analysis according to the x implicit representations of the text sequence to be analyzed obtained by walking the second implicit representation set, and obtaining the target bidding support rate.

4. The method according to claim 3, characterized in that The step of obtaining the confidence of the xth target text sequence corresponding to the sth normal model based on the xth second implicit representation to obtain the adapted confidence includes: obtaining S modified confidences between the xth second implicit representation and the S first normal model coefficients respectively; fusing the S modified confidences to obtain a modified confidence fusion result; obtaining, from the S modified confidences, a target modified confidence corresponding to the first normal model coefficient of the sth normal model and the xth second implicit representation; using the weight of the target modified confidence in the modified confidence fusion result as the adapted confidence; the step of obtaining the confidence of the xth target text sequence corresponding to the sth normal model based on the xth second implicit representation; ... After obtaining the adaptive configuration confidence, the method further comprises: for the first normal model coefficient, cyclically performing the following processing: obtaining the second normal model coefficient of the sth normal model according to the x adaptive configuration confidences of the sth first normal model coefficient obtained by walking the second implicit representation set and the second implicit representation set; iterating the adaptive configuration confidence based on the second normal model coefficient and the xth second implicit representation to obtain the target adaptive configuration confidence; if the iterative processing loop stops, re-representing according to the S target adaptive configuration confidences of the xth second implicit representation and the S second normal model coefficients of the S normal models to obtain the implicit representation of the text sequence to be analyzed.

5. The method according to claim 4, characterized in that The method of obtaining the second normal model coefficient of the sth normal model based on the x adapted confidences of the sth first normal model coefficient obtained by walking the second implicit representation set and the second implicit representation set includes: fusing the x adapted confidences of the sth first normal model coefficient obtained by walking the second implicit representation set to obtain an adapted confidence fusion result; fusing the x adapted confidences with the second implicit representation set to obtain a confidence implicit representation fusion result; and obtaining the second normal model coefficient having a change trend consistent with that of the confidence implicit representation fusion result and opposite to that of the adapted confidence fusion result.

6. The method according to any one of claims 1 to 5, characterized in that Before performing the bidding possibility analysis based on the second implicit representation set to obtain the target bidding support rate, the method also includes: performing activation processing based on the first implicit representation set to obtain an implicit representation significant coefficient set; performing the bidding possibility analysis based on the second implicit representation set to obtain the target bidding support rate includes: merging the second implicit representation set with the implicit representation significant coefficient set to obtain an implicit representation set to be fused; fusing the first implicit representation set and the implicit representation set to be fused to obtain a target implicit representation set; performing the bidding possibility analysis based on the target implicit representation set to obtain the target bidding support rate.

7. The method according to any one of claims 1 to 5, characterized in that The target bid support rate is obtained based on the bid possibility analysis neural network processing, and the acquisition process of the bid possibility analysis neural network includes: performing bid possibility analysis on the bidding document training text by initializing the neural network to obtain the predicted bid support rate; obtaining the target training cost based on the error between the predicted bid support rate and the bid support first-verification mark of the bidding document training text; training the initialized neural network based on the target training cost to obtain the bid possibility analysis neural network.

8. The method according to claim 7, characterized in that The method of training the initialized neural network according to the target training cost to obtain the bid possibility analysis neural network includes: obtaining a third normal model coefficient set iterated from each of the R bidding document training texts to obtain R third normal model coefficient sets, wherein the third normal model coefficient set includes S third normal model coefficients corresponding to S normal models, and R≥2; fusing the R third normal model coefficient sets to determine a normal model coefficient fusion set; combining the initial normal model coefficient set of the bid possibility analysis neural network and the normal model coefficient fusion set to obtain a normal model coefficient iteration set; adjusting the network internal configuration variables of the initialized neural network according to the target training cost to obtain updated network internal configuration variables; combining the normal model coefficient iteration set and the updated network internal configuration variables to obtain the bid possibility analysis neural network.

9. The method according to any one of claims 1 to 5, characterized in that: Based on the target bid support rate, a matching preset bid document generation framework is called to generate a bid document, including: calling a target bid document generation network trained in advance; inputting a general bid text and the matching preset bid document generation framework into the target bid document generation network, and obtaining a bid document based on the target bid document generation network; wherein the training process of the target bid document generation network includes: obtaining a bid training text and an example bid document generation framework; in the bid document generation network to be trained, greedily searching the bid training text through the example bid document generation framework to obtain a reduced dimension representation vector of the training text; in the bid document generation network to be trained, The reduced-dimensional representation vector of the training text is beam searched through the example bidding document generation framework to obtain the increased-dimensional representation vector of the training text, and the reduced-dimensional representation vector of the training text and the increased-dimensional representation vector of the training text are weightedly fused through the initial fusion variable to obtain the training prediction representation vector; a cost function is generated based on the training prediction representation vector and the bidding training text, and the initial fusion variable is optimized based on the cost function until the representation vector fusion variable is obtained when the variable is stable, and the bidding document generation network to be trained including the representation vector fusion variable is determined as the target bidding document generation network; the target bidding document generation network is used to generate a bidding document based on the bidding document generation framework.

10. A computer system, characterized in that: include: one or more processors; Memory; One or more computer programs; wherein the one or more computer programs are stored in the memory and configured to be executed by the one or more processors, and when the one or more computer programs are executed by the processors, the method according to any one of claims 1 to 9 is implemented.

Images