Intelligent bidding method, system, equipment and medium for project material bidding

Abstract

The present invention discloses an intelligent bidding method, system, device and medium for project material bidding. First, semantic representation and preprocessing of text are completed through a preset BERT model, semantic information of text is learned efficiently, and the accuracy of category division is improved; then, the description text is preliminarily classified through the BERT model to obtain a parent category label, and the parent category label is fused with the corresponding description text as the input of the next layer to predict the subcategory label, which effectively improves the accuracy and efficiency of classification; the subcategory label is combined with the description text through the BERT model, and the optimal weight parameter obtained by training the previous layer network is loaded into the current layer network, so that the classification result of the current layer has the characteristics of the classification result of the previous layer label, and the accuracy of category label prediction is improved; finally, the minimum hash algorithm is used to realize the merging of bid sections, so as to solve the problem of bidding difficulties due to insufficient pre-investment amount of the bid section, and improve the success rate of project bidding.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent bid division, and in particular to an intelligent bid division method, system, equipment and medium for project material bidding. Background Art

[0002] At present, economic growth is driving social development to change. In order to adapt to the new trend of integration and innovation of advanced technologies and industries such as artificial intelligence, big data, and blockchain, various industries are actively promoting "Internet + bidding and procurement" and are committed to building and developing electronic bidding and its information integration and sharing system. In the field of bidding, the division of bid sections is the primary task of bidding planning. The bidding plan of the project is directly related to the development of all subsequent bidding-related work. It is also a key component of the bidding documents, providing the necessary basis and basis for bid evaluation and bidding. In the actual bidding of engineering projects, the division of bid sections is a very important link. The division of bid sections has a vital role and influence on the implementation and completion of the entire engineering project. Only scientific, reasonable and legal division of bid sections can ensure the smooth implementation and high-quality completion of engineering projects. Therefore, in the actual bidding process, it is necessary to pay attention to the scientific rationality of the division of bid sections.

[0003] The bidding work of traditional power grid construction project materials is usually done manually, which consumes a lot of manpower and material resources and is prone to errors. Unreasonable bidding schemes will not only hinder the advancement of project construction, but also may cause equipment or site resource conflicts for contractors in different bidding sections, further increasing the difficulty of coordination and management during project construction. Therefore, it is urgent to use artificial intelligence related technologies to realize intelligent bidding, improve the establishment of electronic bidding systems, and promote the reduction of costs and increase of efficiency in bidding business. Summary of the invention

[0004] The present invention aims to solve at least one of the technical problems existing in the prior art. To this end, the present invention proposes an intelligent bidding method, system, device and medium for project material bidding, which can analyze the bidding party's material procurement list, classify the purchased materials into hierarchical categories based on the description text, calculate the similarity of the purchased materials between the bid sections with insufficient pre-investment amounts, merge the relevant bid sections of the same type, and finally realize the bid section division.

[0005] In a first aspect, an embodiment of the present invention provides an intelligent bidding method for project material bidding, including:

[0006] Obtain a material procurement set for each bidding project, where each element in the material procurement set corresponds to a material, and each material has a corresponding description text;

[0007] Input the description text into a preset BERT model to obtain a first feature vector having a parent class label;

[0008] Input the first feature vector into a preset fully connected network to obtain a parent class label classification result corresponding to the parent class label; jointly embed the parent class label classification result and the description text into the BERT model to obtain a second feature vector with a first subclass label; input the second feature vector into the fully connected network to obtain a first subclass label classification result corresponding to the first subclass label; jointly embed the first subclass label classification result and the description text into the BERT model to obtain a third feature vector with a second subclass label; input the third feature vector into the fully connected network to obtain a second subclass label classification result corresponding to the second subclass label; iterate in a loop until the Nth subclass label classification result corresponding to the Nth subclass label is obtained, where N represents a preset subclass label quantity threshold;

[0009] Calculate the pre-investment amount of each bidding section in the Nth sub-category label classification result, and mark the bidding sections whose pre-investment amount is less than the amount threshold as bidding sections to be merged;

[0010] Screening candidate bid lots with the same N-1th subcategory label as each of the bid lots to be merged and having a value greater than or equal to the threshold value;

[0011] The similarity between each of the bid sections to be merged and the Nth subclass label of the corresponding candidate bid section is calculated according to the minimum hash algorithm, and the bid sections are merged according to the similarity to obtain the final bid division result.

[0012] The method according to the embodiment of the present invention has at least the following beneficial effects:

[0013] First, the semantic representation and preprocessing of the text are completed through the preset BERT model. The BERT model can efficiently learn the semantic information of the text and improve the accuracy of material classification. Then, the description text is preliminarily classified through the BERT model to obtain a parent category label. Secondly, the parent category label is fused with the corresponding description text as the input of the next layer to predict the subcategory label. Repeat the iteration until the threshold of the number of subcategory labels is reached. This can effectively improve the accuracy and efficiency of classification, because the classification problem is decomposed into multiple sub-problems through hierarchical labels. Each sub-problem is less difficult and easier to solve. At the same time, it also makes full use of the hierarchical structure of the text and decomposes the category labels of the text into multiple levels, so as to more accurately describe the category characteristics of the text. Most importantly, the subcategory label is combined with the description text through the BERT model. Since there is a certain relationship between the category label and the description text, considering the structural relationship between the category labels can improve the classification effect. The optimal weight parameters obtained by training the previous layer of the network are loaded into this layer of the network, so that the classification results of this layer have the characteristics of the classification results of the previous layer of labels, thereby improving the accuracy of category label prediction. Finally, the minimum hash algorithm is used to realize the merging of bid sections, which can calculate the similarity of the content of purchased materials between bid sections with insufficient pre-investment amounts, merge related bid sections of the same type, and finally realize the division of bid sections, which can effectively solve the problem of difficulty in bidding due to insufficient pre-investment amounts in traditional bidding, and is conducive to improving the success rate of project bidding.

[0014] According to some embodiments of the present invention, the BERT model is composed of a stack of encoders in multiple Transformers, and the encoders are used to perform word embedding, segment embedding and position embedding on the input text.

[0015] According to some embodiments of the present invention, the preset BERT model is obtained by pre-training through the following steps:

[0016] Get unlabeled text data;

[0017] Cutting the unlabeled text data into a plurality of first segments;

[0018] Mark each of the first fragments by covering a portion with a mask to obtain a second fragment;

[0019] The second segment is input into the BERT model for training until the BERT model converges to obtain the preset BERT model; wherein the training process of the BERT model includes executing a tag mask language model task to predict the tags covered by the mask and executing a next sentence prediction task to predict the semantic relationship of the sentence order.

[0020] According to some embodiments of the present invention, the fully connected network is composed of a multilayer perceptron network and a double-layer activation function; the multilayer perceptron network includes an input layer, a hidden layer and an output layer, the hidden layer includes a plurality of neurons, each of the neurons represents a fully connected layer, the first layer of the double-layer activation function is a linear rectification function, and the second layer of the double-layer activation function is a normalized exponential function.

[0021] According to some embodiments of the present invention, the step of calculating the similarity between each of the to-be-merged bid sections and the Nth subclass label of the corresponding candidate bid section according to the minimum hash algorithm comprises the following steps:

[0022] Randomly shuffle the element positions of the Nth subclass label in the bid section to be merged and the candidate bid section by using a hash function to obtain a plurality of first element sets corresponding to the bid section to be merged and a plurality of second element sets corresponding to the candidate bid section;

[0023] Taking the first element in each of the first element sets as the eigenvalue of the first element set to obtain a first characteristic matrix, and taking the first element in each of the second element sets as the eigenvalue of the second element set to obtain a second characteristic matrix;

[0024] Randomly shuffle the first feature matrix and the second feature matrix to obtain a plurality of first minimum hash values ​​corresponding to the first feature matrix and a second minimum hash value corresponding to the second feature matrix;

[0025] The probability of equality between the first minimum Hash value and the second minimum Hash value is calculated to obtain the similarity.

[0026] According to some embodiments of the present invention, calculating the probability of equality between the first minimum Hash value and the second minimum Hash value to obtain the similarity comprises the following steps:

[0027] Using the plurality of the first minimum hash values ​​as a first hash value set, and using the plurality of the second minimum hash values ​​as a second hash value set;

[0028] Calculate the minimum hash equality probability between the first hash value set and the second hash value set by using the Jaccard coefficient;

[0029] The minimum hash equality probability is converted into the similarity.

[0030] According to some embodiments of the present invention, merging the bid sections according to the similarity to obtain a final bid division result includes the following steps:

[0031] Select the candidate bid section with the greatest similarity to obtain the merged bid section corresponding to each bid section to be merged;

[0032] The bid sections to be merged in the Nth sub-category label classification result are merged with the corresponding merged bid sections to obtain a final bid classification result.

[0033] In a second aspect, an embodiment of the present invention provides an intelligent bidding method system for project material bidding, including:

[0034] A material procurement set acquisition unit is used to acquire a material procurement set for each bidding project, wherein each element in the material procurement set corresponds to a material, and each of the materials has a corresponding description text;

[0035] A parent class label classification unit, used for inputting the description text into a preset BERT model to obtain a first feature vector having a parent class label;

[0036] A subclass label classification unit is used to input the first feature vector into a preset fully connected network to obtain a parent class label classification result corresponding to the parent class label; jointly embed the parent class label classification result and the description text into the BERT model to obtain a second feature vector with a first subclass label; input the second feature vector into the fully connected network to obtain a first subclass label classification result corresponding to the first subclass label; jointly embed the first subclass label classification result and the description text into the BERT model to obtain a third feature vector with a second subclass label; input the third feature vector into the fully connected network to obtain a second subclass label classification result corresponding to the second subclass label; and iterate in a loop until an Nth subclass label classification result corresponding to the Nth subclass label is obtained, wherein N represents a preset subclass label quantity threshold;

[0037] A unit for selecting bid sections to be merged, used for calculating the pre-investment amount of each bid section in the Nth sub-category label classification result, and marking the bid sections whose pre-investment amount is less than the amount threshold as bid sections to be merged;

[0038] A merged candidate bidding section selection unit is used to select candidate bidding sections with a value greater than or equal to the amount threshold under the same N-1th sub-category label as each of the bidding sections to be merged;

[0039] The bid segment merging unit is used to calculate the similarity between each bid segment to be merged and the Nth subclass label of the corresponding candidate bid segment according to the minimum hash algorithm, and merge the bid segments according to the similarity to obtain the final bid segmentation result.

[0040] In a third aspect, an embodiment of the present invention provides an electronic device comprising at least one control processor and a memory for communicating with the at least one control processor; the memory stores instructions executable by the at least one control processor, and the instructions are executed by the at least one control processor so that the at least one control processor can execute the intelligent bidding method for project material bidding as described in the first aspect.

[0041] In a fourth aspect, an embodiment of the present invention provides a computer storage medium, wherein the computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute the intelligent bidding method for project material bidding as described in the first aspect.

[0042] It should be noted that the beneficial effects between the second to fourth aspects of the present invention and the prior art are the same as the beneficial effects of the intelligent bidding method for project material bidding of the first aspect, and will not be described in detail here.

[0043] Other features and advantages of the present invention will be set forth in the description which follows, and in part will be apparent from the description, or may be learned by practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0045] Figure 1 This is a flow chart of an intelligent bidding method for project material bidding provided by an embodiment of the present invention;

[0046] Figure 2 is a flowchart of pre-training of a preset BERT model provided by an embodiment of the present invention;

[0047] Figure 3 is a flowchart of calculating the similarity between each to-be-merged bid section and the Nth subclass label of the corresponding candidate bid section according to the minimum hash algorithm provided by an embodiment of the present invention;

[0048] Figure 4 is a flow chart of calculating the probability of equality between a first minimum hash value and a second minimum hash value to obtain similarity provided by an embodiment of the present invention;

[0049] Figure 5 It is a flow chart of obtaining a final bid division result by merging bid sections according to similarity provided by an embodiment of the present invention;

[0050] Figure 6 It is a schematic diagram of an intelligent bidding method for project material bidding provided by an embodiment of the present invention;

[0051] Figure 7 It is a schematic diagram of a process of classification using a BERT model provided in one embodiment of the present invention;

[0052] Figure 8 1 is a schematic diagram of the working principle of the BERT model provided by an embodiment of the present invention;

[0053] Fig. 9 1 is a schematic diagram of the principle of BERT model pre-training provided by an embodiment of the present invention;

[0054] Fig.10 is a structural diagram of a multilayer perceptron provided by an embodiment of the present invention;

[0055] Fig.11 It is a pseudo code schematic diagram of a model training algorithm of a multilayer perceptron based on a back propagation algorithm provided by an embodiment of the present invention;

[0056] Fig.12 is a schematic diagram of an activation function double-layer perceptron structure provided by an embodiment of the present invention;

[0057] Fig.13 is a curve diagram of the change of the sub-standard loss value in a specific embodiment provided by an embodiment of the present invention;

[0058] Fig.14 is a curve diagram showing the variation of the accuracy of the marking in a specific embodiment provided by an embodiment of the present invention;

[0059] Fig.15 It is a structural diagram of an intelligent bidding method system for project material bidding provided by an embodiment of the present invention;

[0060] Fig.16 It is a structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0061] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0062] In the description of the present invention, if there is a description of first, second, etc., it is only for the purpose of distinguishing technical features, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features or implicitly indicating the order of the indicated technical features.

[0063] In the description of the present invention, it should be understood that descriptions involving orientation, such as orientation or positional relationship indicated as up, down, etc., are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on the present invention.

[0064] In the description of the present invention, it should be noted that, unless otherwise clearly defined, terms such as setting, installing, connecting, etc. should be understood in a broad sense, and technicians in the relevant technical field can reasonably determine the specific meanings of the above terms in the present invention in combination with the specific content of the technical solution.

[0065] The technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described below are only some embodiments of the present invention, not all embodiments.

[0066] Reference Figure 1 and Figure 6 In some embodiments of the present invention, a method for intelligent bidding for project material bidding is provided, comprising:

[0067] Step S100: Obtain a material procurement set for each bidding project, where each element in the material procurement set corresponds to a material, and each material has a corresponding description text.

[0068] It should be noted that the description text includes relevant texts such as material name, material number, material description, quantity, budget amount, etc., and can be expanded according to the needs of the bidding task scenario.

[0069] Step S200: Input the description text into a preset BERT model to obtain a first feature vector having a parent class label.

[0070] It should be noted that since this is the first time the preset BERT model is input, you only need to input the description text into the preset BERT model to obtain the first feature vector with the parent class label. The preset BERT model is a pre-trained model, so it can classify the description text into parent class labels more accurately.

[0071] Step S300, input the first feature vector into a preset fully connected network to obtain a parent class label classification result corresponding to the parent class label; jointly embed the parent class label classification result and the description text into the BERT model to obtain a second feature vector with a first subclass label; input the second feature vector into the fully connected network to obtain a first subclass label classification result corresponding to the first subclass label; jointly embed the first subclass label classification result and the description text into the BERT model to obtain a third feature vector with a second subclass label; input the third feature vector into the fully connected network to obtain a second subclass label classification result corresponding to the second subclass label; iterate the loop until the Nth subclass label classification result corresponding to the Nth subclass label is obtained, where N represents the preset subclass label number threshold.

[0072] It should be noted that, refer to Figure 7 For bidding projects of different sizes and types, the requirements for the precision of the sub-classification are different, so the threshold of the number of sub-class labels needs to be determined according to the requirements of the bidding project. At the same time, the BERT model uses the joint embedding method of text and category labels to concatenate the category labels obtained by the previous classification prediction with the corresponding description text, and then uses the BERT model to represent the text feature vector. For example, there is a description text F corresponding to a procurement material. i ={fi 1 、fi 2 , Λfi m}, fi m Represents the mth word of the i-th text; text F i The parent category label predicted at the k-1th layer is q i,k-1 , then the text vector at the kth layer is the concatenation of the parent category label and the description text. For the jth layer, F i The text vector representation is as follows:

[0073]

[0074] Therefore, each description text of purchased materials corresponds to multiple labels, and there is a hierarchical tree structure between these labels, that is, a parent category label can correspond to multiple subcategory labels, and a subcategory label only corresponds to one parent category label. In the material bidding scenario, each material has different description texts and category labels. Since there is a certain relationship between the category label and the description text, considering the structural relationship between the category labels can improve the classification effect. Hierarchical classification is an effective classification method. It loads the optimal weight parameters obtained by training the previous layer of the network into the current layer of the network, so that the classification results of this layer have the characteristics of the classification results of the parent category label of the previous layer. The model concatenates the category label predicted by the i-1 layer with the corresponding text, and extracts features through the BERT model to obtain the joint vector representation of the parent category label and the description text, and then puts it into the category label prediction of the i-th layer of the model, thereby realizing hierarchical label prediction and improving the accuracy of category label prediction.

[0075] Step S400, calculate the pre-investment amount of each bidding section in the Nth sub-category label classification result, and mark the bidding sections whose pre-investment amount is less than the amount threshold as the bidding sections to be merged.

[0076] Step S500: Screen candidate bid lots with an amount greater than or equal to the amount threshold under the same N-1th subcategory label as each bid lot to be merged.

[0077] It should be noted that in order to ensure that the labels of the merged lots are similar and avoid misclassification problems, the lots to be merged are first based on the same N-1th subcategory label, which not only reduces the complexity of subsequent similarity calculations, but also provides the most similar data set for selection for lot merging, avoiding problems such as large data deviation and misclassification after lot merging.

[0078] Step S600: Calculate the similarity between each bid segment to be merged and the Nth subclass label of the corresponding candidate bid segment according to the minimum hash algorithm, and merge the bid segments according to the similarity to obtain the final bid division result.

[0079] This method first completes the semantic representation and preprocessing of the text through the preset BERT model. The BERT model can efficiently learn the semantic information of the text and improve the accuracy of material classification. Then, the description text is preliminarily classified through the BERT model to obtain a parent category label. Secondly, the parent category label is fused with the corresponding description text as the input of the next layer to predict the subcategory label. Repeat the iteration until the threshold of the number of subcategory labels is reached. This can effectively improve the accuracy and efficiency of classification, because the classification problem is decomposed into multiple sub-problems through hierarchical labels, and each sub-problem is less difficult and easier to solve. At the same time, it also makes full use of the hierarchical structure of the text and decomposes the category labels of the text into multiple levels, so as to more accurately describe the category characteristics of the text. Most importantly, the subcategory label is combined with the description text through the BERT model. Since there is a certain relationship between the category label and the description text, considering the structural relationship between the category labels can improve the classification effect. The optimal weight parameters obtained by training the previous layer of the network are loaded into this layer of the network, so that the classification results of this layer have the characteristics of the classification results of the previous layer of labels, thereby improving the accuracy of category label prediction. Finally, the minimum hash algorithm is used to realize the merging of bid sections, which can calculate the similarity of the content of purchased materials between bid sections with insufficient pre-investment amounts, merge related bid sections of the same type, and finally realize the division of bid sections, which can effectively solve the problem of difficulty in bidding due to insufficient pre-investment amounts in traditional bidding, and is conducive to improving the success rate of project bidding.

[0080] In some embodiments of the present invention, the BERT model is composed of a stack of encoders in multiple Transformers, and the encoders are used to perform word embedding, segment embedding, and position embedding on the input text.

[0081] It should be noted that, refer to Figure 8, the BERT model is a language representation model stacked by multiple Transformer Encoders. It uses a bidirectional language model to train a large-scale unlabeled corpus to obtain rich semantic information of the text and generate a bidirectional language representation, that is, it takes into account the context of the text at the same time. Among them, the three tasks of word embedding, segment embedding and position embedding will convert the description text and category label text of the purchased materials into corresponding vectors. For example, referring to the figure, there is a description text "PTN equipment, 80G". The BERT model adds a [CLS] tag at the beginning of the text and a [SEP] tag at the end, and uses the final output vector as the semantic representation of the text. Word embedding uses the WordPiece word segmentation algorithm to query the word vector table and convert each word in the text into a one-dimensional vector; segment embedding automatically learns the global semantic information of the text by querying the segment vector table and merges it with the semantic information of a single word to generate a segment vector. The vector is used to determine whether each word belongs to sentence A or sentence B (assuming that the segment vector has only two types, A and B); position embedding converts its position information into a position vector by querying the position vector table according to the position difference of each word in the text and the semantic information it carries. Finally, the word vector, segment vector and position vector are superimposed as the input vector of the model to capture the semantic information of the description text for subsequent pre-training tasks.

[0082] The BERT model consists of multiple stacked encoders and completes three tasks: word embedding, segment embedding, and position embedding for the input text. It converts the description text and category label text of purchased materials into corresponding vectors, which can better combine the semantic information brought by the label text and description text, and further improve the accuracy of the model's semantic analysis.

[0083] Reference Figure 2 In some embodiments of the present invention, the preset BERT model is pre-trained by the following steps:

[0084] Step S201: Obtain unlabeled text data.

[0085] Step S202: cutting the unlabeled text data into a plurality of first segments.

[0086] Step S203: mask a portion of each first segment to obtain a second segment.

[0087] Step S204: input the second segment into the BERT model for training until the BERT model converges to obtain a preset BERT model; wherein the training process of the BERT model includes executing a tag mask language model task to predict the tag covered by the mask and executing a next sentence prediction task to predict the semantic relationship of the sentence order.

[0088] It should be noted that, refer to Fig. 9In order to further improve the accuracy of the model's semantic analysis, the BERT model pre-trains the Transformer's Encoder network, predicts text masking information and word order relationships, and learns the semantic changes of the text context. For example, referring to the figure, in the pre-training stage, the BERT model uses large-scale unlabeled text data for training. First, the text is divided into multiple continuous segments, and a special mask (i.e., the MASK tag in the figure) is used in each segment to cover part of the tags; secondly, the Masked Language Model task is performed. The BERT model predicts the masked part of the tags in the context without knowing the word position; at the same time, the Next Sentence Prediction task is performed to train the sentence pairs with correct and incorrect word orders in the text. The model recognizes each sentence and the corresponding sentence order in order to model the semantic relationship, that is, randomly input a sentence pair A and B, complete the next sentence prediction task, and predict whether sentence B is the next sentence of sentence A. Finally, the semantic relationship analysis of the description text and the category label text is completed to prepare for the subsequent text classification task.

[0089] Through the pre-training of the BERT model, we can first respond to the classification of parent category labels more accurately, ensuring a good data premise for the subsequent classification of subcategory labels. Then, through the label mask language model task and the next sentence prediction task, we can predict the text masking information and word order relationship, learn the semantic changes of the text context, and provide a good model foundation for the intelligent labeling method, which is convenient for the subsequent data processing of label text and description text, and improves data processing efficiency.

[0090] In some embodiments of the present invention, the fully connected network is composed of a multilayer perceptron network and a double-layer activation function; the multilayer perceptron network includes an input layer, a hidden layer and an output layer, the hidden layer includes multiple neurons, each neuron represents a fully connected layer, the first layer of the double-layer activation function is a linear rectification function, and the second layer of the double-layer activation function is a normalized exponential function.

[0091] It should be noted that, refer to Fig.10 , Multilayer Perceptron (MLP) is a feedforward neural network implemented using BP (Back Propagation, BP) neural network. The MLP structure diagram is as follows Figure 5As shown. The input layer of MLP transfers the input data information to the next layer, and the neurons in the hidden layer are connected to the neurons in the previous layer, so it can be considered that each neuron in the hidden layer has passed through a fully connected layer. The hidden layer is mainly responsible for the nonlinear transformation and feature extraction tasks of the input data, extracting features from the joint embedding vector of the description text of the purchased materials and the parent category label, and stacking multiple fully connected layers to learn and represent deeper feature information, thereby enhancing the model's ability to process and express data. The output layer is mainly responsible for outputting the classification results with the help of mapping the features extracted by the hidden layer. Figure 5 The input variables are X=x[0], x[1], x[2], x[3], W represents the connection coefficient, b represents the bias, y represents the output, function f represents the activation function, and the output of the hidden layer is f(WX+b). The activation function is a very important component in the neural network model, which can give the neural network model nonlinear characteristics and further improve the model's ability to fit nonlinear problems and other complex problems. The input value in the MLP generates the output value through the activation function, and then the back propagation algorithm is used to train the model, adjust the error between the output value of the objective function and the actual output value of the function, and modify the various weight parameters of the intermediate layer. The process is iterated continuously until the error between the output value of the model and the output value of the objective function reaches the minimum value. Model training algorithms based on the back propagation algorithm are as follows: Fig.11 shown.

[0092] The choice of activation function has a great influence on the performance of the neural network model. In practical applications, choosing a reasonable activation function can effectively improve the performance and accuracy of the neural network model, thereby better solving practical problems. The activation function of this embodiment adopts a double-layer perceptron structure. The double-layer perceptron structure diagram is shown in FIG. Fig.12 As shown in the figure, the activation function used in the first layer is the rectified linear unit (ReLU), which transforms the neural network into an approximator of nonlinear functions through nonlinear transformation to improve the network's expressive power and enable the model to learn more complex properties and laws. 1 represents the connection coefficient from the input layer to the hidden layer, b 1 Represents the bias, and the ReLU function calculation formula is as follows:

[0093] r = ReLU(W 1 x i +b 1 );

[0094] The second layer activation function is the normalized exponential function (softmax function for short), which is used to convert the result of each output classification into a corresponding probability value, so as to obtain the label classification result of the i-th layer through probability analysis, W 2represents the connection coefficient from the hidden layer to the output layer, b 2 Represents the bias, and the softmax function calculation formula is as follows:

[0095] c i =softmax(W 2 x i +b 2 );

[0096] By stacking multiple fully connected layers through a multi-layer perceptron and extracting features from the joint embedding vector of the description text and category labels of the purchased materials, deeper feature information can be learned and represented, thereby enhancing the model's ability to process and express data. Then, the linear rectification function and normalized exponential function of the double-layer perceptron structure are used to improve the network's expressive power, allowing the model to learn more complex properties and laws.

[0097] Reference Figure 3 In some embodiments of the present invention, the similarity between each to-be-merged bid segment and the Nth subclass label of the corresponding candidate bid segment is calculated according to the minimum hash algorithm, including the following steps:

[0098] Step S610: Randomly shuffle the element positions of the Nth subclass labels in the bid section to be merged and the candidate bid section by using a hash function to obtain a plurality of first element sets corresponding to the bid section to be merged and a plurality of second element sets corresponding to the candidate bid section;

[0099] Step S620: taking the first element in each first element set as the eigenvalue of the first element set to obtain a first characteristic matrix, and taking the first element in each second element set as the eigenvalue of the second element set to obtain a second characteristic matrix;

[0100] Step S630: Randomly shuffle the first characteristic matrix and the second characteristic matrix to obtain a plurality of first minimum hash values ​​corresponding to the first characteristic matrix and a second minimum hash value corresponding to the second characteristic matrix.

[0101] Step S640: Calculate the probability of equality between the first minimum hash value and the second minimum hash value to obtain similarity.

[0102] It should be noted that the minimum hash algorithm uses a random hash function to map sets to signatures and calculates the similarity between sets with reference to the signatures. Specifically, first, use a hash function to randomly and uniformly shuffle the element positions of the Nth subclass labels in the corresponding bid sections to be merged and candidate bid sections to obtain multiple first element sets corresponding to the bid sections to be merged and multiple second element sets corresponding to the candidate bid sections; the first element of each set is regarded as the eigenvalue of the set to form a feature matrix, which contains the position information of the elements in the set; secondly, by shuffling the matrix multiple times, each set generates multiple minimum hash values, and the column vectors are converted into MinHash signatures to form a signature matrix, which represents the characteristics of the set; finally, the probability that the minimum hash values ​​of the sets are equal is used to represent the similarity.

[0103] The Min-Hash algorithm is used to analyze the similarities between similar procurement materials, and multiple bidding sections with similar content and small pre-investment amounts are merged to further ensure the rationality of the bidding results.

[0104] Reference Figure 4 In some embodiments of the present invention, calculating the probability of equality between the first minimum hash value and the second minimum hash value to obtain the similarity includes the following steps:

[0105] Step S641: Use multiple first minimum hash values ​​as a first hash value set, and use multiple second minimum hash values ​​as a second hash value set.

[0106] Step S642: Calculate the minimum hash equality probability between the first hash value set and the second hash value set by using the Jaccard coefficient.

[0107] Step S643: Convert the minimum hash equality probability into similarity.

[0108] It should be noted that the Jaccard coefficient calculates the similarity between sets by comparing the intersection and union of two sets. Suppose there are set A and set B, the number of elements in the intersection is X, and the number of elements in the union is X+Y. The calculation formula of the Jaccard coefficient similarity, that is, the probability that the minimum hash values ​​are equal, is as follows:

[0109]

[0110] It provides a more accurate Jaccard coefficient estimate and has a smaller variance and higher accuracy.

[0111] Reference Figure 5 In some embodiments of the present invention, merging bid sections according to similarity to obtain a final bid division result includes the following steps:

[0112] Step S601: Select the candidate bid section with the greatest similarity to obtain the merged bid section corresponding to each bid section to be merged.

[0113] Step S602: merge the bid sections to be merged in the Nth sub-category label classification result with the corresponding merged bid sections to obtain the final bid division result.

[0114] In order to ensure that there are no errors in the material categories, only candidate bidding sections with the greatest similarity are selected to ensure the rationality of the bidding results.

[0115] In order to verify the effectiveness of this method, a specific experimental embodiment of an intelligent bidding method for project material bidding is provided, including:

[0116] Regarding experimental data, this embodiment uses a total material data set from the material bidding of power grid construction projects, as shown in Table 1. Each material data in the data set includes material description, major category description (label), medium category description (label), minor category description (label), etc. For the convenience of the experiment, this embodiment only uses three labels, and the three labels are obtained through the above-mentioned intelligent bid classification method for project material bidding. In order to improve the generalization ability, reduce the model's excessive attention to certain material data features, and reduce the risk of overfitting, the training data needs to be randomly shuffled.

[0117] Table 1 Material data set

[0118]

[0119] The model in this embodiment uses hierarchical multi-label classification to classify material descriptions, and uses deep learning to train the model. Due to the large amount of data involved, the model needs to be trained using a GPU. The operating system of this experiment is Windows 11, the programming language used is Python 3.9, the development tool is Pycharm 2023, the graphics card is NVIDIA GeForceRTX 4060Laptop GPU, the model framework is Pytorch-gpu==2.0.0+cu118, and the processor is AMD Ryzen 7 7840Hw / Radeon 780M Graphics (16CPUS).

[0120] The experimental parameter information is set on the total material data set of the material bidding for the power grid construction project. The experimental parameters are shown in Table 2.

[0121] Table 2 Experimental parameters

[0122]

[0123] For the hierarchical material category label classification of the intelligent classification model proposed in this embodiment, the general accuracy (Accuracy) is selected, W1 represents the accuracy of material major category classification, W2 represents the accuracy of material medium category classification, W3 represents the accuracy of material small category classification, precision (Precision), recall rate (Recall), F1 value (F1-Score) is selected as evaluation indicators.

[0124] Reference Fig.13 During the model training process, the model loss on the training set and the loss on the test set gradually increase, and the overfitting phenomenon of the model worsens. In order to solve this problem, the early stopping strategy is used to observe the training process of the model. The main idea is that when it is found that the test error is larger than the test error obtained last time, the test will not be terminated immediately, but training and testing will continue for several epochs. If the test error still does not decrease, it is considered that the experiment stopped when the lowest test error was reached last time. This strategy saves the parameters of the model during training when the model is overfitting, and improves the generalization ability of the model. Fig.13 The middle figure (a) shows the case where the stopping strategy is not used, and the figure (b) shows the case where the stopping strategy is not used.

[0125] Experimental results and comparative analysis: To verify the superiority of the hierarchical model for the experimental data set of this embodiment, a common planar MTC model is used for comparison. There is no obvious hierarchical structure between the labels in the data set of the planar MTC task, and each data may belong to one or more labels. TextCNN is a relatively outstanding planar model for processing MTC tasks. The model uses CNN and uses pre-trained word vectors as text features. The performance comparison results of the model of this embodiment and TextCNN are shown in Table 3.

[0126] Table 3 Performance comparison between the model in this embodiment and the TextCNN model

[0127]

[0128] On the total data set of experimental materials, the overall accuracy of the model is used, that is, the accuracy of the model in the last layer of classification. This is because when each layer is classified, the real label of the previous layer is not provided to the next layer, and the label obtained by the next layer comes from the label predicted by the previous layer. The ordinary planar multi-label algorithm does not process the intermediate layers in the hierarchy, and only considers the classification of the last layer. The classification performance of the model in this embodiment is better than TextCNN, with Accuracy increased by 26.97%, Precision increased by 26.88%, Recall increased by 28.74%, and F1-Score increased by 27.82%. Compared with Text-CNN for multi-label text classification, the model predicts category labels by layer-by-layer analysis, reflecting better classification performance. This is because the BERT model can not only extract the feature representation of a polysemous word, but also calculate the weight of each position relative to another position in parallel without relying on the order between words, thereby enhancing the prediction performance of the word vector model.

[0129] The intelligent label classification model proposed in this embodiment adopts fine-tuning technology. In the BERT pre-training stage, the model is further trained by fine-tuning BERT using data related to specific fields or tasks. In order to further study the impact of fine-tuning on the label prediction performance of the model, the model of this embodiment is compared with the model without fine-tuning BERT, and the performance comparison results are shown in Table 4.

[0130] Table 4 Comparison of label prediction performance between the model in this embodiment and the model without fine-tuning

[0131]

[0132] From the results in Table 4 and the principle of fine-tuning, we can see that because the fine-tuning target domain has a similar network structure to the source domain, the correlation between labels can be used to classify the material description category through fine-tuning technology, that is, the model loads the optimal parameters obtained by training the upper network into the network layer to achieve prediction, which can effectively improve the accuracy.

[0133] Ablation analysis is a technique used to understand and evaluate neural network models. It studies the impact of certain components or functions in the neural network on model performance by removing them. The model in this embodiment uses ablation to remove hierarchical label embedding for classification, and only uses BERT to classify the purchased materials into category labels to obtain large category labels, medium category labels, and small category labels. The performance comparison results of the ablation experiment of the model in this embodiment are shown in Table 5. The accuracy change curve of the model in this embodiment is shown in Fig.14 shown.

[0134] Table 5 Performance comparison of the model in this embodiment and the model ablation experiment

[0135]

[0136] Depend on Fig.14 As shown in Table 5, the model of this embodiment shows good performance in category label prediction, and compared with the ablation model that only uses BERT and a fully connected network and removes hierarchical label embedding, the accuracy of the first-layer category label prediction of the model of this embodiment has not changed significantly because label embedding is not used. By embedding the first-layer labels, the second-layer category label classification accuracy is improved by 2.97%, and by embedding the second-layer labels, the third-layer category label classification accuracy is improved by 7.09%, and the overall classification accuracy is improved by 7.06%. It can be seen that the classification model proposed in this embodiment has stronger performance, and it also shows that the hierarchical idea has exploratory significance for improving the accuracy of classification.

[0137] Reference Fig.15 An embodiment of the present invention further provides an intelligent bidding method system for project material bidding, comprising a material procurement set acquisition unit 1001, a parent label classification unit 1002, a sub-category label classification unit 1003, a bid section selection unit 1004 to be merged, a merged candidate bid section selection unit 1005 and a bid section merging unit 1006, wherein:

[0138] The material procurement set acquisition unit 1001 is used to acquire the material procurement set of each bidding project. Each element in the material procurement set corresponds to a material, and each material has a corresponding description text.

[0139] The parent class label classification unit 1002 is used to input the description text into a preset BERT model to obtain a first feature vector with a parent class label.

[0140] The subclass label classification unit 1003 is used to input the first feature vector into a preset fully connected network to obtain a parent class label classification result corresponding to the parent class label; jointly embed the parent class label classification result and the description text into the BERT model to obtain a second feature vector with a first subclass label; input the second feature vector into the fully connected network to obtain a first subclass label classification result corresponding to the first subclass label; jointly embed the first subclass label classification result and the description text into the BERT model to obtain a third feature vector with a second subclass label; input the third feature vector into the fully connected network to obtain a second subclass label classification result corresponding to the second subclass label; and iterate in a loop until the Nth subclass label classification result corresponding to the Nth subclass label is obtained, where N represents a preset subclass label quantity threshold.

[0141] The bidding section selection unit 1004 to be merged is used to calculate the pre-investment amount of each bidding section in the Nth sub-category label classification result, and mark the bidding sections whose pre-investment amount is less than the amount threshold as the bidding sections to be merged.

[0142] The merged candidate bidding section selection unit 1005 is used to screen candidate bidding sections with a monetary value greater than or equal to a monetary value threshold under the same N-1th subcategory label as each bidding section to be merged.

[0143] The bid segment merging unit 1006 is used to calculate the similarity between each bid segment to be merged and the Nth subclass label of the corresponding candidate bid segment according to the minimum hash algorithm, and merge the bid segments according to the similarity to obtain the final bid segmentation result.

[0144] It should be noted that since the intelligent bidding method system for project material bidding in this embodiment and the above-mentioned intelligent bidding method for project material bidding are based on the same inventive concept, the corresponding contents in the method embodiment are also applicable to the present device embodiment and will not be described in detail here.

[0145] refer to Fig.16 Another embodiment of the present invention further provides an electronic device, and the electronic device 6000 can be any type of smart terminal, such as a mobile phone, a tablet computer, a personal computer, etc.

[0146] Specifically, the electronic device 6000 includes: one or more control processors 6001 and a memory 6002, Fig.16 In the example, a control processor 6001 and a memory 6002 are used. The control processor 6001 and the memory 6002 can be connected via a bus or other means. Fig.16 The example of connecting through bus is taken in the following.

[0147] The memory 6002 is a non-transitory computer-readable storage medium that can be used to store non-transitory software programs, non-transitory computer executable programs and modules, such as program instructions / modules corresponding to an electronic device in an embodiment of the present invention;

[0148] The control processor 6001 executes various functional applications and data processing of an intelligent bidding method for project material bidding by running non-transient software programs, instructions and modules stored in the memory 6002, that is, an intelligent bidding method for project material bidding that implements the above-mentioned method embodiment.

[0149] The memory 6002 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and an application required by at least one function; the data storage area may store data created by the use of an intelligent bidding method for project material bidding, etc. In addition, the memory 6002 may include a high-speed random access memory, and may also include a non-transitory memory, such as at least one disk storage device, a flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory 6002 may optionally include a memory remotely arranged relative to the control processor 6001, and these remote memories may be connected to the electronic device 6000 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0150] In one or more modules stored in the memory 6002, when executed by the one or more control processors 6001, an intelligent bidding method for project material bidding in the above method embodiment is executed, for example, the above described Figures 1 to 5 method steps.

[0151] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0152] It should be noted that since an electronic device in this embodiment and the above-mentioned intelligent bidding method for project material bidding are based on the same inventive concept, the corresponding contents in the method embodiment are also applicable to the device embodiment and will not be described in detail here.

[0153] One embodiment of the present invention further provides a computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions are used to execute: an intelligent bidding method for project material bidding as in the above embodiment.

[0154] It should be noted that since a computer-readable storage medium in this embodiment and the above-mentioned intelligent bidding method for project material bidding are based on the same inventive concept, the corresponding contents in the method embodiment are also applicable to the device embodiment and will not be described in detail here.

[0155] It will be appreciated by those skilled in the art that all or some of the steps and systems in the disclosed method above may be implemented as software, firmware, hardware and appropriate combinations thereof. Some physical components or all physical components may be implemented as software executed by a processor, such as a central processing unit, a digital signal processor or a microprocessor, or may be implemented as hardware, or may be implemented as an integrated circuit, such as an application specific integrated circuit. Such software may be distributed on a computer-readable medium, which may include a computer storage medium (or a non-transitory medium) and a communication medium (or a temporary medium). As known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing data (such as computer-readable instructions, data structures, program modules or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, disk storage or other magnetic storage devices, or any other medium that may be used to store desired data and may be accessed by a computer. Furthermore, it is well known to those skilled in the art that communication media typically embodies computer readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any data delivery media.

[0156] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "illustrative embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0157] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

Claims

1. An intelligent bidding method for project material bidding, characterized in that: The intelligent bidding method for project material bidding includes: Obtain a material procurement set for each bidding project, where each element in the material procurement set corresponds to a material, and each material has a corresponding description text; Input the description text into a preset BERT model to obtain a first feature vector having a parent class label; Input the first feature vector into a preset fully connected network to obtain a parent class label classification result corresponding to the parent class label; jointly embed the parent class label classification result and the description text into the BERT model to obtain a second feature vector with a first subclass label; input the second feature vector into the fully connected network to obtain a first subclass label classification result corresponding to the first subclass label; jointly embed the first subclass label classification result and the description text into the BERT model to obtain a third feature vector with a second subclass label; input the third feature vector into the fully connected network to obtain a second subclass label classification result corresponding to the second subclass label; iterate in a loop until the Nth subclass label classification result corresponding to the Nth subclass label is obtained, where N represents a preset subclass label quantity threshold; Calculate the pre-investment amount of each bidding section in the Nth sub-category label classification result, and mark the bidding sections whose pre-investment amount is less than the amount threshold as bidding sections to be merged; Screening candidate bid lots with the same N-1th subcategory label as each of the bid lots to be merged and having a value greater than or equal to the threshold value; The similarity between each of the bid sections to be merged and the Nth subclass label of the corresponding candidate bid section is calculated according to the minimum hash algorithm, and the bid sections are merged according to the similarity to obtain the final bid division result.

2. The intelligent bidding method for project material bidding according to claim 1 is characterized in that: The BERT model consists of a stack of encoders in multiple Transformers, which are used to embed word embedding, segment embedding, and position embedding for the input text.

3. The intelligent bidding method for project material bidding according to claim 2 is characterized in that: The preset BERT model is pre-trained by the following steps: Get unlabeled text data; Cutting the unlabeled text data into a plurality of first segments; Mark each of the first fragments by covering a portion with a mask to obtain a second fragment; The second segment is input into the BERT model for training until the BERT model converges to obtain the preset BERT model; wherein the training process of the BERT model includes executing a tag mask language model task to predict the tags covered by the mask and executing a next sentence prediction task to predict the semantic relationship of the sentence order.

4. The intelligent bidding method for project material bidding according to claim 1 is characterized in that: The fully connected network is composed of a multilayer perceptron network and a double-layer activation function; the multilayer perceptron network includes an input layer, a hidden layer and an output layer, the hidden layer includes a plurality of neurons, each of the neurons represents a fully connected layer, the first layer of the double-layer activation function is a linear rectification function, and the second layer of the double-layer activation function is a normalized exponential function.

5. The intelligent bidding method for project material bidding according to claim 1 is characterized in that: The calculating the similarity between each of the to-be-merged bid sections and the Nth subclass label of the corresponding candidate bid section according to the minimum hash algorithm comprises the following steps: Randomly shuffle the element positions of the Nth subclass label in the bid section to be merged and the candidate bid section by using a hash function to obtain a plurality of first element sets corresponding to the bid section to be merged and a plurality of second element sets corresponding to the candidate bid section; Taking the first element in each of the first element sets as the eigenvalue of the first element set to obtain a first characteristic matrix, and taking the first element in each of the second element sets as the eigenvalue of the second element set to obtain a second characteristic matrix; Randomly shuffle the first feature matrix and the second feature matrix to obtain a plurality of first minimum hash values ​​corresponding to the first feature matrix and a second minimum hash value corresponding to the second feature matrix; The probability of equality between the first minimum Hash value and the second minimum Hash value is calculated to obtain the similarity.

6. The intelligent bidding method for project material bidding according to claim 5 is characterized in that: The calculating the probability of equality between the first minimum Hash value and the second minimum Hash value to obtain the similarity comprises the following steps: Using the plurality of the first minimum hash values ​​as a first hash value set, and using the plurality of the second minimum hash values ​​as a second hash value set; Calculate the minimum hash equality probability between the first hash value set and the second hash value set by using the Jaccard coefficient; The minimum hash equality probability is converted into the similarity.

7. The intelligent bidding method for project material bidding according to claim 6 is characterized in that: The step of merging the bid sections according to the similarity to obtain the final bid division result comprises the following steps: Select the candidate bid section with the greatest similarity to obtain the merged bid section corresponding to each bid section to be merged; The bid sections to be merged in the Nth sub-category label classification result are merged with the corresponding merged bid sections to obtain a final bid classification result.

8. An intelligent bidding method system for project material bidding, characterized in that: The intelligent bidding method system for project material bidding includes: A material procurement set acquisition unit is used to acquire a material procurement set for each bidding project, wherein each element in the material procurement set corresponds to a material, and each of the materials has a corresponding description text; A parent class label classification unit, used for inputting the description text into a preset BERT model to obtain a first feature vector having a parent class label; A subclass label classification unit is used to input the first feature vector into a preset fully connected network to obtain a parent class label classification result corresponding to the parent class label; jointly embed the parent class label classification result and the description text into the BERT model to obtain a second feature vector with a first subclass label; input the second feature vector into the fully connected network to obtain a first subclass label classification result corresponding to the first subclass label; jointly embed the first subclass label classification result and the description text into the BERT model to obtain a third feature vector with a second subclass label; input the third feature vector into the fully connected network to obtain a second subclass label classification result corresponding to the second subclass label; and iterate in a loop until an Nth subclass label classification result corresponding to the Nth subclass label is obtained, wherein N represents a preset subclass label quantity threshold; A unit for selecting bid sections to be merged, used for calculating the pre-investment amount of each bid section in the Nth sub-category label classification result, and marking the bid sections whose pre-investment amount is less than the amount threshold as bid sections to be merged; A merged candidate bidding section selection unit is used to select candidate bidding sections with a value greater than or equal to the amount threshold under the same N-1th sub-category label as each of the bidding sections to be merged; The bid segment merging unit is used to calculate the similarity between each bid segment to be merged and the Nth subclass label of the corresponding candidate bid segment according to the minimum hash algorithm, and merge the bid segments according to the similarity to obtain the final bid segmentation result.

9. An electronic device, characterized in that: It includes at least one control processor and a memory for communicating with the at least one control processor; the memory stores instructions that can be executed by the at least one control processor, and the instructions are executed by the at least one control processor so that the at least one control processor can execute the intelligent bidding method for project material bidding as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute the intelligent bidding method for project material bidding as described in any one of claims 1 to 7.

Images