System for helping bidding and pricing analysis based on tender information big data

Abstract

The application discloses a system for bid price analysis based on tender information big data, and relates to the technical field of intelligent decision-making. A tender information acquisition module is used for acquiring current and historical tender information full data. A semantic analysis module is connected with the tender information acquisition module and is used for extracting the implicit tendency and price sensitivity characteristics of a tender document, and generating a tender tendency semantic feature vector. A competitor portrait module is connected with the tender information acquisition module and is used for constructing a competitive product dynamic pricing probability distribution. A pricing game simulation module is connected with the semantic analysis module and the competitor portrait module and is used for generating a winning probability curve through counterfactual game simulation. The semantic analysis module is used for carrying out clause-level coding and attention weighting on the tender document, quantitatively extracting the tendency deviation degree and the price sensitivity elasticity coefficient, and converting the implicit tendency into a calculable semantic feature vector, so that the explicit modeling of the influence of non-price clauses on the business tender score is realized.

Description

Technical Field

[0001] This invention relates to the field of intelligent decision-making technology, and in particular to a system for assisting in bid pricing analysis based on big data of bidding information. Background Technology

[0002] In open bidding activities, the reasonableness of the bid price directly determines the probability of winning the bid and the project's profit margin. Existing bid price analysis methods mainly have the following shortcomings: Most systems rely on historical average bid prices or cost-plus pricing as the core basis, treating bid prices as a single numerical variable without delving into the implicit biases contained in the tender documents. Tender documents often contain statements that exclude low prices, guide the prioritization of quality, and tolerate budget deviations. These non-price clauses substantially influence the formation of the benchmark price for bid evaluation and the calculation of the commercial bid score, but existing systems lack structured extraction and quantitative modeling of the semantics of the text.

[0003] Analysis of competitors' pricing behavior remains at the level of static averages. Competitors' pricing strategies change dynamically with market conditions, project types, and time. There are significant differences between projects of different sizes, in different regions, and with different technical requirements. Simple historical averages cannot reflect the drift trend of pricing distribution, resulting in a disconnect between the predicted distribution and the actual competitive landscape of the current project.

[0004] The methods for estimating the probability of winning the bid are rather crude. Some systems use deterministic ranking, which directly compares scores after assuming competitors' bids are fixed; others use linear regression to fit the probability of winning the bid. None of these methods can answer the counterfactual question of what the outcome would be if our side adopted a specific bid, and they are unable to characterize the non-linear relationship between bid and probability of winning the bid.

[0005] The bidding optimization process lacks risk constraints. Simply pursuing maximum expected profit may result in bids falling into a range with extremely low probability of winning the bid; once the bid is lost, the profit is zero or even sunk costs are incurred. The existing system does not incorporate risk metrics such as conditional value of risk into the optimization objective, making it impossible to achieve a balance between returns and risks.

[0006] Therefore, there is an urgent need for a bid pricing analysis system that can integrate the semantics of bidding texts, dynamically model competitors' bidding behavior, support counterfactual game simulation, and have risk constraints. Summary of the Invention

[0007] To address the aforementioned technical problems, this invention provides a system for analyzing bid prices based on big data from bidding information. The technical solution adopted is as follows: A system that leverages big data from bidding and tendering information to assist in bid pricing analysis includes: The tender information acquisition module is used to collect all current and historical tender information data. The semantic parsing module, connected to the tender information acquisition module, is used to extract implicit bias and price sensitivity features from tender documents and generate a semantic feature vector of tender bias. The competitor profiling module, connected to the bidding information collection module, is used to construct the dynamic bidding probability distribution of competitors. The bidding game simulation module connects to the semantic parsing module and the competitor profiling module, and is used to generate a winning probability curve for counterfactual game simulation. The pricing optimization module, connected to the pricing game simulation module, is used to solve for the optimal pricing range under risk constraints. The output module connects to the quotation optimization module and is used to output standardized quotation suggestions. The system executes a method to assist in bid pricing analysis, including the following steps: Step 1: Collect historical and current bidding information data through the bidding information collection module; Step 2: Extract the semantic feature vector of bidding tendency through the semantic parsing module; Step 3: Construct the probability distribution of competitor pricing behavior and calculate the pricing drift coefficient using the competitor profiling module; Step 4: Perform counterfactual game simulation using the bidding game simulation module to generate a simulated winning probability curve; Step 5: Solve for the optimal price range using the price optimization module; Step 6: Output a quotation suggestion through the output module.

[0008] Optionally, step 2, which involves extracting the semantic feature vector of bidding preference, can be done as follows: The current tender document text is segmented into clauses, and invalid paragraphs are removed. Each clause is encoded using a pre-trained language model to obtain a clause-level semantic vector. The semantic vectors of clauses containing price, technical deviation, and commercial deviation are subjected to attention-weighted pooling to obtain a tender bias semantic base vector. Bid bias words and negative structures in the clauses are identified, and the bias deviation degree δ and price sensitivity elasticity coefficient λ are calculated. The tender bias semantic base vector, bias deviation degree δ, and price sensitivity elasticity coefficient λ are concatenated to form a tender bias semantic feature vector.

[0009] Optionally, step 3 includes the following sub-steps: Step 31: Filter out similar projects from historical bidding information data that have a similarity greater than a threshold with the current bidding information data; Step 32: For each competitor, calculate the ratio r of its historical bids relative to the tender budget price; Step 33: Construct the probability distribution of competitors' bidding behavior using Gaussian kernel density estimation. ; Step 34: Using the time decay factor and project similarity as weights, perform a weighted average of the deviations in the historical ratio r to obtain the competitor's bid drift coefficient in the current tender. ;

[0010] Step 35, adjust the price drift coefficient. Translate and overlay the probability distribution of the competitor's bidding behavior Obtain the predicted distribution of competitor quotes .

[0011] Optionally, in step 34, the quote drift coefficient The calculation formula is: ; : The bid drift coefficient of the j-th competitor in the current tender, representing the systematic deviation of the competitor's bidding behavior relative to the historical mean; N is the total number of historical bid records for similar projects after screening; It is the joint weight of the i-th record. ; It is a time coefficient. It is the time interval between the i-th historical record and the current tender. It is a project similarity function. It is the feature vector of the i-th historical item. It is the feature vector of the current bidding project; It is the competitor's bid ratio in the i-th record, defined as the ratio of the competitor's bid to the tender budget price. It is the arithmetic mean of all historical bid ratios of competitors.

[0012] Optionally, in step 4, a candidate quote set is set. For each candidate offer ,Will Substitute our bid into the counterfactual game model; sample M groups of competitor bid combinations from the predicted competitor bid distribution; for each sample, calculate our bid score and competitor bid scores using the bidding tendency semantic feature vector; calculate the proportion of samples where our bid score is higher than all competitor bid scores, and use these as candidate bids. The simulated probability of winning the bid is calculated, and a simulated probability curve of winning the bid is generated.

[0013] Optional, our commercial bid score The calculation formula is: ; Our candidate quote The commercial bid score is as follows: The maximum score for the commercial bid is the total score specified in the tender documents. This is the price deviation penalty coefficient, which represents the severity of the penalty when the bid deviates from the benchmark price for evaluation. This is our k-th candidate offer. It is the benchmark price for bid evaluation or the price preferred by the tendering party, determined by the benchmark price calculation method specified in the tender documents. It is the strength coefficient of the price-sensitive elasticity coefficient, which characterizes the weight of the price-sensitive elasticity's influence on the score. This is the price sensitivity elasticity coefficient, extracted from the tender documents by the semantic parsing module, which characterizes the tenderer's sensitivity to price changes. It is the adjustment strength coefficient for the degree of bias deviation, representing the magnitude of the adjustment of implicit bias on the commercial bid score. It is the bias deviation, which is calculated by the semantic parsing module after identifying bias words and negative structures in the bidding documents. A positive value indicates that the bidder prefers a high price, and a negative value indicates that the bidder prefers a low price.

[0014] Optionally, step 5 includes the following sub-steps: Step 51, with expected profit Let be the objective function, where For cost, Please provide our candidate quotes. This represents the simulated probability of winning the bid when our bid is p. Step 52, set risk constraints: within the bidding range where the simulated winning probability curve is below the safety threshold, calculate the conditional value of risk (CVaR) and require that CVaR not exceed the preset risk limit. Step 53: Within the set of candidate bids that satisfy the risk constraints, search for the optimal bid p that maximizes the objective function Ep; Step 54: Using p as the center, select the neighborhood where the simulated probability of winning the bid decreases by no more than a preset proportion, as the optimal bidding range.

[0015] Optionally, in step 4, the counterfactual game simulation method is as follows: for any candidate bid... Only by changing our own price, while keeping the competitor's price prediction distribution unchanged, we can form a forecast based on repeated sampling if our price is... The counterfactual result; the winning probability of the counterfactual result is a frequency estimate of multiple samplings.

[0016] Optionally, the implicit bias features in the semantic parsing module include at least one of price bias statements, quality bias statements, and budget deviation bias statements; Price bias statements are wording that rejects or encourages low or high prices; quality bias statements are wording that prioritizes quality or cost-effectiveness; budget deviation bias statements are wording that allows or prohibits exceeding the budget price; implicit bias features are weighted to the bidding bias semantic feature vector through attention weights.

[0017] Optionally, the output module outputs a pricing suggestion including the optimal price p*, the optimal price range, and the corresponding simulated winning probability curve; when the width of the optimal price range is less than a preset value, a competitive situation prompt is triggered; the competitive situation prompt is used to indicate the degree of bidding competition and pricing strategy suggestions, including at least one of text, graphics, or color indicators.

[0018] In summary, the present invention has at least one of the following beneficial technical effects: This invention provides a system for analyzing bid prices based on big data from bidding information. Through a semantic parsing module, it performs clause-level coding and attention weighting on bidding documents, quantifies and extracts the bias deviation degree and price sensitivity elasticity coefficient, and transforms implicit biases into calculable semantic feature vectors, thereby achieving explicit modeling of the impact of non-price clauses on the commercial bid score.

[0019] The bid drift coefficient is calculated by weighting the time decay factor and project similarity together, and then the drift coefficient is shifted and superimposed on the bid behavior distribution obtained by Gaussian kernel density estimation. This allows the competitor bid prediction distribution to adapt to the current project characteristics and time dimension, thereby improving the timeliness and targeting of bid behavior prediction.

[0020] By using counterfactual game simulation, while keeping the predicted distribution of competitors' bids unchanged, we repeatedly sample and score our candidate bids to generate a continuous simulated winning probability curve. This supports counterfactual inference of the winning probability for any bid level, overcoming the limitations of deterministic ranking and linear regression.

[0021] By using expected profit as the objective function and conditional risk value not exceeding the risk ceiling as the constraint, the optimal bid and its neighborhood are solved. This approach aims to maximize profits while controlling tail risk in low-probability bid intervals, and outputs a bid range recommendation that is both profitable and robust. Attached Figure Description

[0022] Figure 1 This is a flowchart illustrating the bidding price analysis method implemented by the system based on bidding information big data, according to the present invention. Figure 2 This is a simulated probability curve of winning a bid in a specific embodiment of the present invention. Detailed Implementation

[0023] The present invention will be further described in detail below with reference to the accompanying drawings.

[0024] This invention discloses a system for assisting in bid pricing analysis based on big data from bidding information.

[0025] Reference Figure 1 and Figure 2 Example 1, a system for assisting bid pricing analysis based on bidding information big data, includes: The tender information acquisition module is used to collect all current and historical tender information data. The semantic parsing module, connected to the tender information acquisition module, is used to extract implicit bias and price sensitivity features from tender documents and generate a semantic feature vector of tender bias. The competitor profiling module, connected to the bidding information collection module, is used to construct the dynamic bidding probability distribution of competitors. The bidding game simulation module connects to the semantic parsing module and the competitor profiling module, and is used to generate a winning probability curve for counterfactual game simulation. The pricing optimization module, connected to the pricing game simulation module, is used to solve for the optimal pricing range under risk constraints. The output module connects to the quotation optimization module and is used to output standardized quotation suggestions. The system executes a method to assist in bid pricing analysis, including the following steps: Step 1: Collect historical and current bidding information data through the bidding information collection module; Step 2: Extract the semantic feature vector of bidding tendency through the semantic parsing module; Step 3: Construct the probability distribution of competitor pricing behavior and calculate the pricing drift coefficient using the competitor profiling module; Step 4: Perform counterfactual game simulation using the bidding game simulation module to generate a simulated winning probability curve; Step 5: Solve for the optimal price range using the price optimization module; Step 6: Output a quotation suggestion through the output module.

[0026] By adopting the above technical solution, the tender information acquisition module is used to collect all current and historical tender information data. The semantic parsing module, connected to the tender information acquisition module, is used to extract implicit biases and price-sensitive features from the tender documents, generating a semantic feature vector of tender biases. The competitor profiling module, connected to the tender information acquisition module, is used to construct a dynamic probability distribution of competitor bids. The bidding game simulation module, connected to both the semantic parsing module and the competitor profiling module, is used for counterfactual game simulation to generate a winning bid probability curve. The bidding optimization module, connected to the bidding game simulation module, is used to solve for the optimal bidding range under risk constraints. The output module, connected to the bidding optimization module, is used to output standardized bidding suggestions.

[0027] The system executes a method to assist in bid price analysis, which includes steps 1 to 6 in sequence. Step 1 collects historical and current bidding information data through the bidding information collection module.

[0028] Step 2: Extract the semantic feature vector of bidding tendency through the semantic parsing module.

[0029] Step 3: Construct the probability distribution of competitor pricing behavior and calculate the pricing drift coefficient using the competitor profiling module.

[0030] Step 4: Perform counterfactual game simulation using the bidding game simulation module to generate a simulated winning probability curve.

[0031] Step 5 uses the pricing optimization module to find the optimal pricing range.

[0032] Step 6 outputs a quotation suggestion through the output module.

[0033] After the bidding information acquisition module obtains the raw data, the semantic parsing module and the competitor profiling module extract features from the two dimensions of bidding text and competitor historical behavior, respectively. The bidding game simulation module integrates these two types of features to perform counterfactual inference. The bidding optimization module completes the bidding decision under the premise of controllable risk. The output module presents the results in a standardized manner.

[0034] Example 2, the method for extracting the semantic feature vector of bidding tendency in step 2 is as follows: The current tender document text is segmented into clauses, and invalid paragraphs are removed. Each clause is encoded using a pre-trained language model to obtain a clause-level semantic vector. The semantic vectors of clauses containing price, technical deviation, and commercial deviation are subjected to attention-weighted pooling to obtain a tender bias semantic base vector. Bid bias words and negative structures in the clauses are identified, and the bias deviation degree δ and price sensitivity elasticity coefficient λ are calculated. The tender bias semantic base vector, bias deviation degree δ, and price sensitivity elasticity coefficient λ are concatenated to form a tender bias semantic feature vector.

[0035] By adopting the above technical solution, the current tender document text is first segmented into clauses, and invalid paragraphs are removed. Then, a pre-trained language model is used to encode each clause, resulting in a clause-level semantic vector. Next, attention-weighted pooling is applied to the semantic vectors of clauses containing price, technical deviation, and commercial deviation to obtain the tender bias semantic base vector. Subsequently, bias words and negative structures in the clauses are identified, and the bias deviation degree δ and price sensitivity elasticity coefficient λ are calculated. Finally, the tender bias semantic base vector, bias deviation degree δ, and price sensitivity elasticity coefficient λ are concatenated to form the tender bias semantic feature vector.

[0036] Pre-trained language models can capture implicit tendencies in the semantics of clauses, attention-weighted pooling highlights the impact of key clauses, and the tendency deviation and price sensitivity elasticity coefficient quantify the tenderer's specific preferences for price and quality. The combination of these three forms a computable feature vector, providing text-based input for subsequent bidding games.

[0037] Example 3, step 3 includes the following sub-steps: Step 31: Filter out similar projects from historical bidding information data that have a similarity greater than a threshold with the current bidding information data; Step 32: For each competitor, calculate the ratio r of its historical bids relative to the tender budget price; Step 33: Construct the probability distribution of competitors' bidding behavior using Gaussian kernel density estimation. ; Step 34: Using the time decay factor and project similarity as weights, perform a weighted average of the deviations in the historical ratio r to obtain the competitor's bid drift coefficient in the current tender. ; Step 35, adjust the price drift coefficient. Translate and overlay the probability distribution of the competitor's bidding behavior Obtain the predicted distribution of competitor quotes .

[0038] By adopting the above technical solution, step 31 filters out similar projects from historical bidding information data that have a similarity greater than a threshold with the current bidding information data.

[0039] Step 32: For each competitor, calculate the ratio r of its historical bid to the tender budget price.

[0040] Step 33: Construct the probability distribution of competitor bidding behavior using Gaussian kernel density estimation. .

[0041] Step 34 uses the time decay factor and project similarity as weights to perform a weighted average of the deviation of the historical ratio r, and obtains the bid drift coefficient Δj of the competitor in the current tender.

[0042] Step 35: Shift and superimpose the bid drift coefficient Δj onto the competitor's bid behavior probability distribution fjr to obtain the competitor's bid prediction distribution. .

[0043] Screening for similar projects ensures sample relevance; Gaussian kernel density estimation extracts a continuous distribution from historical bid ratios; time decay and project similarity weighting enable the bid drift coefficient to reflect the latest changes in competitor behavior and project suitability; translation and superposition form a dynamic prediction distribution, improving the accuracy of bid prediction.

[0044] Example 4, step 34, price drift coefficient The calculation formula is: ; : The bid drift coefficient of the j-th competitor in the current tender, representing the systematic deviation of the competitor's bidding behavior relative to the historical mean; N is the total number of historical bid records for similar projects after screening; It is the joint weight of the i-th record. ; It is a time coefficient. It is the time interval between the i-th historical record and the current tender. It is a project similarity function. It is the feature vector of the i-th historical item. It is the feature vector of the current bidding project; It is the competitor's bid ratio in the i-th record, defined as the ratio of the competitor's bid to the tender budget price. It is the arithmetic mean of all historical bid ratios of competitors.

[0045] By adopting the above technical solution, in step 34, the formula for calculating the price drift coefficient Δj is as follows: Where Δj is the bid drift coefficient of the j-th competitor in the current tender, representing the systematic deviation of the competitor's bidding behavior relative to the historical mean. N is the total number of historical bid records for similar projects after filtering. It is the joint weight of the i-th record. Where τ is the time coefficient, is the time interval between the i-th historical record and the current tender, and sim is the project similarity function. It is the feature vector of the i-th historical item. It is the feature vector of the current bidding project. It is the competitor's bid ratio in the i-th record, defined as the ratio of the competitor's bid to the tender budget price. It is the arithmetic mean of all historical bid ratios of competitors. The working principle is: weighting. Taking into account both time decay and project similarity, recent records with high similarity contribute more to the drift coefficient, enabling Δj to adapt to short-term changes in competitors' pricing strategies and differences in project characteristics, thus overcoming the lag of simple historical means.

[0046] Example 5, in step 4, a candidate quote set is set. For each candidate offer ,Will Substitute our bid into the counterfactual game model; sample M groups of competitor bid combinations from the predicted competitor bid distribution; for each sample, calculate our bid score and competitor bid scores using the bidding tendency semantic feature vector; calculate the proportion of samples where our bid score is higher than all competitor bid scores, and use these as candidate bids. The simulated probability of winning the bid is calculated, and a simulated probability curve of winning the bid is generated.

[0047] By adopting the above technical solution, in step 4, the candidate price set P is first set to include... , until For each candidate offer ,Will This is used as the basis for a counterfactual game model. M groups of competitor bid combinations are sampled from the predicted competitor bid distribution. For each sample, the scores of our bid and competitors' bids are calculated using the semantic feature vector of bidding tendency. The proportion of samples where our bid score is higher than all competitors' bid scores is used as the simulated probability of winning the bid in the candidate bid competition, thus generating a simulated probability curve. The working principle is: by fixing the competitor distribution and only changing our bid, a counterfactual inference is formed; multiple samplings simulate different competitive scenarios, and the probability of winning is estimated by frequency, avoiding the bias of a single deterministic prediction; at the same time, the bidding text tendency is incorporated into the score calculation, so that the probability of winning reflects the true evaluation logic.

[0048] Example 6, Our Commercial Bid Score The calculation formula is: ; Our candidate quote The commercial bid score is as follows: The maximum score for the commercial bid is the total score specified in the tender documents. This is the price deviation penalty coefficient, which represents the severity of the penalty when the bid deviates from the benchmark price for evaluation. This is our k-th candidate offer. It is the benchmark price for bid evaluation or the price preferred by the tendering party, determined by the benchmark price calculation method specified in the tender documents. It is the strength coefficient of the price-sensitive elasticity coefficient, which characterizes the weight of the price-sensitive elasticity's influence on the score. This is the price sensitivity elasticity coefficient, extracted from the tender documents by the semantic parsing module, which characterizes the tenderer's sensitivity to price changes. It is the adjustment strength coefficient for the degree of bias deviation, representing the magnitude of the adjustment of implicit bias on the commercial bid score. It is the bias deviation, which is calculated by the semantic parsing module after identifying bias words and negative structures in the bidding documents. A positive value indicates that the bidder prefers a high price, and a negative value indicates that the bidder prefers a low price.

[0049] By adopting the above technical solution, our commercial bid score was [high / low]. The calculation formula is: ; in, Our candidate quote The business bid score is as follows. The commercial bid benchmark score is the maximum score, calculated as the total score stipulated in the tender documents. α is the price deviation penalty coefficient, representing the severity of the penalty when the bid deviates from the benchmark price. pk is our k-th candidate bid. This is the benchmark price or the bidder's preferred price, determined by the benchmark price calculation method specified in the bidding documents. β is the strength coefficient of the price sensitivity elasticity coefficient, representing the weight of the price sensitivity elasticity on the score. λ is the price sensitivity elasticity coefficient, extracted from the bidding documents by the semantic parsing module, representing the bidder's sensitivity to price changes. μ is the correction strength coefficient of the bias deviation, representing the adjustment range of implicit bias on the commercial bid score. δ is the bias deviation, calculated by the semantic parsing module after identifying bias words and negative structures in the bidding documents; a positive value indicates the bidder's preference for a high price, and a negative value indicates a preference for a low price. The working principle is: this formula, based on the traditional price deviation deduction, introduces two corrections—the price sensitivity elasticity coefficient and the bias deviation—to ensure that the bid score considers both price competition and the bidder's implicit preferences, thus more accurately simulating the scoring behavior of evaluation experts.

[0050] Example 7, step 5 includes the following sub-steps: Step 51, with expected profit Let be the objective function, where For cost, Please provide our candidate quotes. This represents the simulated probability of winning the bid when our bid is p. Step 52, set risk constraints: within the bidding range where the simulated winning probability curve is below the safety threshold, calculate the conditional value of risk (CVaR) and require that CVaR not exceed the preset risk limit. Step 53: Within the set of candidate bids that satisfy the risk constraints, search for the optimal bid p that maximizes the objective function Ep; Step 54: Using p as the center, select the neighborhood where the simulated probability of winning the bid decreases by no more than a preset proportion, as the optimal bidding range.

[0051] By adopting the above technical solution, step 51 aims to achieve the desired profit. Let c be the objective function, where c is the cost and p is our candidate bid. This represents the simulated probability of winning the bid when our bid is p.

[0052] Step 52 sets risk constraints: within the bidding range where the simulated winning probability curve is below the safety threshold, calculate the conditional value of risk (CVaR), requiring that CVaR not exceed a preset risk upper limit. Sub-step 53 searches within the set of candidate bids that satisfy the risk constraints to make the objective function... The largest optimal price p.

[0053] Step 54: Using p as the center, select a neighborhood where the simulated probability of winning the bid decreases by no more than a preset percentage, as the optimal bidding range.

[0054] The expected profit maximization ensures the commercial objective of the bid, the CVaR constraint controls the tail loss risk in the low probability of winning the bid scenario, and the neighborhood selection takes into account both profit and stability, so that the final bid recommendation has operability and risk buffering capacity.

[0055] In Example 8, step 4, the method for counterfactual game simulation is as follows: for any candidate bid... Only by changing our own price, while keeping the competitor's price prediction distribution unchanged, we can form a forecast based on repeated sampling if our price is... The counterfactual result; the winning probability of the counterfactual result is a frequency estimate of multiple samplings.

[0056] By adopting the above technical solution, any candidate price can be... Only by changing our own price, while keeping the competitor's price prediction distribution unchanged, we can form a forecast based on repeated sampling if our price is... The counterfactual results are derived from the frequency estimates of the winning probability obtained through multiple sampling. The working principle is as follows: the counterfactual framework simulates the causal effect of a single variable change, avoiding confounding effects caused by simultaneous changes in multiple variables; the frequency estimation uses the law of large numbers to approximate the true winning probability, giving the simulation results statistical reliability.

[0057] Example 9: The implicit bias features in the semantic parsing module include at least one of price bias statements, quality bias statements, and budget deviation bias statements; Price bias statements are wording that rejects or encourages low or high prices; quality bias statements are wording that prioritizes quality or cost-effectiveness; budget deviation bias statements are wording that allows or prohibits exceeding the budget price; implicit bias features are weighted to the bidding bias semantic feature vector through attention weights.

[0058] By adopting the above technical solution, the implicit bias features in the semantic parsing module include at least one of price bias statements, quality bias statements, and budget deviation bias statements. Price bias statements are phrases that reject or encourage low or high prices; quality bias statements are phrases that prioritize quality or cost-effectiveness; and budget deviation bias statements are phrases that allow or prohibit exceeding the budget. These implicit bias features are weighted into the bidding bias semantic feature vector using attention weights.

[0059] By defining three types of bias statements, the system can extract non-explicit clauses from bidding documents in a structured manner. The attention weight is automatically adjusted according to the importance of the clauses, making the bias features both comprehensive and highlighting the key points, thus making up for the shortcomings of traditional methods that only focus on price clauses.

[0060] Example 10: The output module outputs a bidding suggestion including the optimal bid p*, the optimal bid range, and the corresponding simulated winning probability curve; when the width of the optimal bid range is less than a preset value, a competitive situation prompt is triggered; the competitive situation prompt is used to indicate the degree of bidding competition and bidding strategy suggestions, including at least one of text, graphics, or color indicators.

[0061] By adopting the above technical solution, the output module provides pricing suggestions including the optimal price p*, the optimal price range, and the corresponding simulated probability curve of winning the bid. When the width of the optimal price range is less than a preset value, a competitive situation alert is triggered. The competitive situation alert indicates the sufficiency of bidding competition and provides pricing strategy suggestions, including at least one of text, graphics, or color indicators.

[0062] The width of the optimal bid range reflects the intensity of competition; the smaller the width, the narrower the bidding space and the more intense the competition. Triggering prompts can help bidders quickly identify risks and adjust their strategies. Multiple prompt formats are available to suit different use cases, improving decision-making efficiency.

[0063] The following describes the implementation principle of the present invention using specific embodiments:

[0064] A company participated in a public tender for a smart park information technology construction project. The tender budget was 10 million yuan. The tender documents stipulated that the commercial bid had a maximum score of 60 points, and the price score was calculated as the arithmetic average of valid bids minus 5% as the benchmark price. Our cost (c) was 7.2 million yuan. Six similar projects with a similarity greater than 0.8 were identified from historical tenders, involving competitors A, B, and C.

[0065] Step 1: Data Collection; The tender information collection module collects the full text of the current tender documents, the tender budget price, our costs, and the winning bid announcements and quotation records of similar projects in the past three years. The collection results include price preference statements, quality preference statements, and budget deviation preference statements in the tender documents, as well as the historical quotation ratios and time information of competitors A, B, and C in six similar projects.

[0066] Step 2: Semantic feature extraction results; The semantic parsing module segments and semantically encodes the tender documents. After attention-weighted pooling and bias word identification, a price sensitivity elasticity coefficient λ=0.42 and a bias deviation δ=0.15 are obtained, indicating that the tenderer has a slight preference for high-quality and high-priced solutions. Commercial bid benchmark score. =60 points, price deviation penalty coefficient α=0.8, price sensitivity elasticity effect strength β=0.35, tendency deviation correction strength μ=1.2.

[0067] Step 3: Competitor profiling and bid drift calculation;

[0068] For competitor A, the bid ratio based on 6 historical records Number of days since today and project similarity As shown in Table 1: Table 1 1 0.920 50 0.631 2 0.900 100 0.608 3 0.935 200 0.611 4 0.890 300 0.608 5 0.960 450 0.627 6 0.885 600 0.601 Time coefficient =0.001 Arithmetic mean =0.915. The formula for calculating the joint weight is: The weighted sum is 2.8299. Deviation. = The formula for calculating the price drift factor is: ;

[0069] Substituting the values, we get ΔA = 0.00262. Similarly, we calculate ΔB = −0.018 for B and ΔC = 0.035 for C. Using Gaussian kernel density estimation, we construct the price ratio distributions for A, B, and C. Then, by shifting and superimposing their respective drift coefficients, we obtain the competitor's price prediction distribution. .

[0070] Step 4: Counterfactual game simulation and winning probability curve; Set a set of candidate offers The price ranges from 7.2 million to 10 million yuan, with each 100,000 yuan increment representing a point. The simulation runs M=5000 times. Candidate bids are used... Taking a price of 8.6 million yuan as an example, a sample yielded bids of 9.35 million yuan from bidder A, 8.8 million yuan from bidder B, and 9.1 million yuan from bidder C. The average effective bid was: =860+935+880+9104=8.9625 million yuan; Benchmark Price for Bidding = ×0.95=8.514375 million yuan; The formula for calculating our commercial bid score is as follows: ; Substituting, we get: =60−0.8×∣860−851.4375∣+0.35×0.42×(860−851.4375)+1.2×0.15; Calculated ≈54.59 points. Competitor scores are calculated using the same formula and taken as non-negative values. Our score was the highest in this sample, and we are determined to win the bid. Repeating the sampling 5000 times, the winning frequency is statistically analyzed to obtain a simulated winning probability of 8.6 million yuan for the candidate bid. =0.42. The simulation probability curve for winning the bid was obtained by repeating the calculation for all candidate bids. Partial results are as follows: 0.61 for 7.6 million yuan, 0.58 for 7.8 million yuan, 0.54 for 8 million yuan, 0.50 for 8.2 million yuan, 0.46 for 8.4 million yuan, 0.42 for 8.6 million yuan, 0.37 for 8.8 million yuan, 0.31 for 9 million yuan, 0.25 for 9.2 million yuan, 0.19 for 9.4 million yuan, 0.13 for 9.6 million yuan, 0.08 for 9.8 million yuan, and 0.04 for 10 million yuan.

[0071] Step 5: Solve for the optimal price range by considering expected profit and risk constraints; Cost c = 7.2 million yuan. For each candidate bid... Calculate expected profit: #imgpt130#; The calculations yield the following expected profits: 7.6 million yuan: 40 × 0.61 = 24.4; 40 × 0.61 = 24.4 million yuan; 7.8 million yuan: 34.8 million yuan; 8 million yuan: 43.2 million yuan; 8.2 million yuan: 50.0 million yuan; 8.4 million yuan: 55.2 million yuan; 8.6 million yuan: 58.8 million yuan; 8.8 million yuan: 59.2 million yuan; 9 million yuan: 55.8 million yuan; 9.2 million yuan: 50.0 million yuan; 9.4 million yuan: 41.8 million yuan; 9.6 million yuan: 31.2 million yuan; 9.8 million yuan: 20.8 million yuan; 10 million yuan: 11.2 million yuan. The maximum expected profit occurs at 8.8 million yuan, at 59.2 million yuan. The safety threshold for the probability of winning the bid is set at 0.20, and the upper limit of the Conditional Value at Risk (CVaR) is 30 million yuan. The expected loss for unsuccessful bids of 9.4 million yuan and above exceeds 30 million yuan, thus failing to meet the risk constraints. Therefore, the optimal bid is 8.8 million yuan. Using 8.8 million yuan as the center, the simulated probability of winning the bid decreases by no more than a preset percentage of 10%, corresponding to a winning probability of no less than 0.333. The bid range is approximately 8.4 million yuan to 9 million yuan, with the optimal bid range being 8.4 million yuan to 9 million yuan.

[0072] Step 6: Output the results; The output module outputs the optimal bid price #imgpt132#=8.8 million yuan, with the optimal bid price range being 8.4 million yuan to 9 million yuan, and simulates the probability curve of winning the bid. Figure 2 The final output of the pricing optimization module is displayed in the form of an information panel, including the optimal bid of 8.8 million yuan, the optimal bid range of 8.4 million to 9 million yuan, an index explanation of the simulated winning probability curve, and competitive situation information. The panel uses large fonts and clear borders, making it easy to read directly from the accompanying diagrams in the manual. This diagram corresponds to the output of step 6, intuitively presenting the standard pricing suggestions generated by the system and demonstrating the standardized information display function of the output module.

[0073] The optimal price range is 600,000 yuan, which is greater than the preset value of 200,000 yuan. The competitive situation indicator is that the level of competition is moderate, and there is no need to trigger a strong prompt.

[0074] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A system for analyzing bid prices based on big data from bidding and tendering information, characterized in that: include: The tender information acquisition module is used to collect all current and historical tender information data. The semantic parsing module, connected to the tender information acquisition module, is used to extract implicit bias and price sensitivity features from tender documents and generate a semantic feature vector of tender bias. The competitor profiling module, connected to the bidding information collection module, is used to construct the dynamic bidding probability distribution of competitors. The bidding game simulation module connects to the semantic parsing module and the competitor profiling module, and is used to generate a winning probability curve for counterfactual game simulation. The pricing optimization module, connected to the pricing game simulation module, is used to solve for the optimal pricing range under risk constraints. The output module connects to the quotation optimization module and is used to output standardized quotation suggestions. The system executes a method to assist in bid pricing analysis, including the following steps: Step 1: Collect historical and current bidding information data through the bidding information collection module; Step 2: Extract the semantic feature vector of bidding tendency through the semantic parsing module; Step 3: Construct the probability distribution of competitor pricing behavior and calculate the pricing drift coefficient using the competitor profiling module; Step 4: Perform counterfactual game simulation using the bidding game simulation module to generate a simulated winning probability curve; Step 5: Solve for the optimal price range using the price optimization module; Step 6: Output a quotation suggestion through the output module.

2. The system for assisting bid pricing analysis based on bidding information big data as described in claim 1, characterized in that, Step 2 involves extracting the semantic feature vector of bidding preference using the following method: The current tender document text is segmented into clauses, and invalid paragraphs are removed. Each clause is encoded using a pre-trained language model to obtain a clause-level semantic vector. The semantic vectors of clauses containing price, technical deviation, and commercial deviation are subjected to attention-weighted pooling to obtain a tender bias semantic base vector. Bid bias words and negative structures in the clauses are identified, and the bias deviation degree δ and price sensitivity elasticity coefficient λ are calculated. The tender bias semantic base vector, bias deviation degree δ, and price sensitivity elasticity coefficient λ are concatenated to form a tender bias semantic feature vector.

3. The system for assisting bid pricing analysis based on bidding information big data as described in claim 2, characterized in that, Step 3 includes the following sub-steps: Step 31: Filter out similar projects from historical bidding information data that have a similarity greater than a threshold with the current bidding information data; Step 32: For each competitor, calculate the ratio r of its historical bids relative to the tender budget price; Step 33: Construct the probability distribution of competitors' bidding behavior using Gaussian kernel density estimation. ; Step 34: Using the time decay factor and project similarity as weights, perform a weighted average of the deviations in the historical ratio r to obtain the competitor's bid drift coefficient in the current tender. ; Step 35, adjust the price drift coefficient. Translate and overlay the probability distribution of the competitor's bidding behavior Obtain the predicted distribution of competitor quotes .

4. The system for assisting bid pricing analysis based on bidding information big data as described in claim 3, characterized in that, In step 34, the price drift coefficient The calculation formula is: ； : The bid drift coefficient of the j-th competitor in the current tender, representing the systematic deviation of the competitor's bidding behavior relative to the historical mean; N is the total number of historical bid records for similar projects after screening; It is the joint weight of the i-th record. ; It is a time coefficient. It is the time interval between the i-th historical record and the current tender. It is a project similarity function. It is the feature vector of the i-th historical item. It is the feature vector of the current bidding project; It is the competitor's bid ratio in the i-th record, defined as the ratio of the competitor's bid to the tender budget price. It is the arithmetic mean of all historical bid ratios of competitors.

5. The system for assisting bid pricing analysis based on bidding information big data as described in claim 4, characterized in that, In step 4, a candidate quote set is set. For each candidate offer ,Will Substitute our bid into the counterfactual game model; sample M groups of competitor bid combinations from the competitor bid prediction distribution; for each sample, calculate our commercial bid score and the competitor's commercial bid score using the bidding tendency semantic feature vector; We will statistically analyze the percentage of bids where our commercial bid score is higher than that of all competitors, and use these percentages as candidate bids. The simulated probability of winning the bid is calculated, and a simulated probability curve of winning the bid is generated.

6. The system for assisting bid pricing analysis based on bidding information big data as described in claim 5, characterized in that, Our commercial bid score The calculation formula is: ； Our candidate quote The commercial bid score is as follows: The maximum score for the commercial bid is the total score specified in the tender documents. This is the price deviation penalty coefficient, which represents the severity of the penalty when the bid deviates from the benchmark price for evaluation. This is our k-th candidate offer. It is the benchmark price for bid evaluation or the price preferred by the tendering party, determined by the benchmark price calculation method specified in the tender documents. It is the strength coefficient of the price-sensitive elasticity coefficient, which characterizes the weight of the price-sensitive elasticity's influence on the score. This is the price sensitivity elasticity coefficient, extracted from the tender documents by the semantic parsing module, which characterizes the tenderer's sensitivity to price changes. It is the adjustment strength coefficient for the degree of bias deviation, representing the magnitude of the adjustment of implicit bias on the commercial bid score. It is the bias deviation, which is calculated by the semantic parsing module after identifying bias words and negative structures in the bidding documents. A positive value indicates that the bidder prefers a high price, and a negative value indicates that the bidder prefers a low price.

7. The system for assisting bid pricing analysis based on bidding information big data as described in claim 6, characterized in that, Step 5 includes the following sub-steps: Step 51, with expected profit Let be the objective function, where For cost, Please provide our candidate quotes. This represents the simulated probability of winning the bid when our bid is p. Step 52, set risk constraints: within the bidding range where the simulated winning probability curve is below the safety threshold, calculate the conditional value of risk (CVaR) and require that CVaR not exceed the preset risk limit. Step 53: Within the set of candidate bids that satisfy the risk constraints, search for the optimal bid p that maximizes the objective function E(p); Step 54: Using p as the center, select the neighborhood where the simulated probability of winning the bid decreases by no more than a preset proportion, as the optimal bidding range.

8. The system for assisting bid pricing analysis based on bidding information big data as described in claim 7, characterized in that, In step 4, the counterfactual game simulation method is as follows: for any candidate bid... Only by changing our own price, while keeping the competitor's price prediction distribution unchanged, we can form a forecast based on repeated sampling if our price is... The counterfactual result; the winning probability of the counterfactual result is a frequency estimate of multiple samplings.

9. The system for assisting bid pricing analysis based on bidding information big data as described in claim 8, characterized in that, The implicit bias features in the semantic parsing module include at least one of price bias statements, quality bias statements, and budget deviation bias statements; Price bias statements are wording that rejects or encourages low or high prices; quality bias statements are wording that prioritizes quality or cost-effectiveness; budget deviation bias statements are wording that allows or prohibits exceeding the budget price; implicit bias features are weighted to the bidding bias semantic feature vector through attention weights.

10. The system for assisting bid pricing analysis based on bidding information big data as described in claim 9, characterized in that, The output module outputs a quotation suggestion including the optimal quotation p*, the optimal quotation range, and the corresponding simulated winning probability curve. When the width of the optimal quotation range is less than a preset value, a competitive situation prompt is triggered. The competitive situation prompt is used to indicate the degree of bidding competition and quotation strategy suggestions, including at least one of text, graphics, or color indicators.

Images