Vehicle price interval prediction method and related products

Abstract

The application provides a vehicle price interval prediction method and related products. The method comprises the following steps: obtaining a target SKU, a potential customer set corresponding to the target SKU, and a plurality of source price signals corresponding to each potential customer. The corresponding plurality of source price signals are aggregated by first quantile on a user granularity to obtain a potential customer willingness-to-pay price corresponding to each potential customer. According to the potential customer willingness-to-pay price corresponding to each potential customer, a sample set of SKU willingness-to-pay prices of the target SKU is constructed. The sample set of SKU willingness-to-pay prices is aggregated by second quantile on a SKU granularity to obtain a lower limit of a guide price interval and an upper limit of the guide price interval. According to the lower limit of the guide price interval and the upper limit of the guide price interval, a guide price interval of the target SKU is generated. The application can reduce the influence of abnormal price signals and the difference in the number of price signals of different potential customers on the prediction result, and improve the stability and accuracy of the guide price interval.

Description

Technical Field

[0001] The embodiments of this application relate to the field of data processing technology, specifically to a method for predicting vehicle price ranges and related products. Background Technology

[0002] With the digital development of vehicle trading, business systems such as vehicle trading platforms and dealer management systems often need to analyze vehicle prices based on historical business data to provide a reference for vehicle pricing and sales management.

[0003] Existing vehicle price analysis methods typically rely primarily on historical transaction data, statistically analyzing the historical prices of similar or identical vehicle models to arrive at a corresponding reference price range. However, vehicle models, configurations, and sales regions vary considerably, and some vehicles have a limited number of historical transactions. This limited transaction data is insufficient to fully reflect consumers' actual acceptance of vehicle prices, easily leading to a lack of stability in the resulting price range.

[0004] During vehicle transactions, interaction records such as clicks, favorites, contact information, and expressions of interest are generated. These records can reflect consumers' interest in vehicles at different price points. However, a single consumer may generate multiple interaction records, and the data volume and price distribution of different interaction records may vary. If these records are directly mixed and statistically analyzed, duplicate records or abnormal prices may significantly affect the statistical results, causing the boundaries of the obtained price range to shift or become too wide.

[0005] Therefore, existing vehicle price analysis methods suffer from insufficient historical transaction data and the problem that directly mixing multiple interaction records can easily lead to distorted results, making it difficult to obtain a stable and practically valuable range of vehicle prices. Summary of the Invention

[0006] The embodiments of this application propose a method for predicting vehicle price ranges and related products.

[0007] Firstly, this application provides a method for predicting vehicle price ranges, the method comprising: Obtain the target SKU, the potential customer set corresponding to the target SKU, and the multi-source price signal corresponding to each potential customer in the potential customer set, wherein the multi-source price signal includes price signals from at least two different sources; For each potential customer, the first quantile of the corresponding multi-source price signals is aggregated at the user granularity to obtain the potential customer's willingness to pay price. Based on the potential customer's willingness to pay price, construct a sample set of SKU willingness to pay prices for the target SKU, wherein the sample set of SKU willingness to pay prices includes at least one potential customer willingness to pay price; At the SKU granularity, the second quantile aggregation is performed on the sample set of the SKU's willingness to pay price to obtain the lower limit and upper limit of the guidance price range for the target SKU; The guidance price range for the target SKU is generated based on the lower limit and the upper limit of the guidance price range.

[0008] In some optional implementations, the step of performing first quantile aggregation on the corresponding multi-source price signals at the user granularity to obtain the potential customer's willingness to pay price includes: Arrange the multi-source price signals corresponding to the potential customers in ascending order of price values ​​to obtain an ordered price signal sequence; Based on the preset lower truncation quantile and upper truncation quantile, the lower quantile threshold and upper quantile threshold of the ordered price signal sequence are determined respectively; Retain price signals in the ordered price signal sequence whose price values ​​are not lower than the lower quantile threshold and not higher than the upper quantile threshold; The median of the retained price signals is determined as the potential customer's willingness to pay price.

[0009] In some optional implementations, the step of performing first quantile aggregation on the corresponding multi-source price signals at the user granularity to obtain the potential customer's willingness to pay price includes: Obtain the signal source type and generation time of each price signal corresponding to the relevant potential customer, as well as the preset source credibility determination rules, exponential time decay method and target weighted quantile; The source reliability of each price signal is determined based on the signal source type and the source reliability determination rules. The time interval corresponding to each price signal is determined based on the generation time of each price signal and the current calculation time. Using the exponential time decay method, the source reliability of each price signal is time decayed according to the time interval corresponding to each price signal to obtain the signal recording weight of each price signal. Arrange the multi-source price signals corresponding to the potential customers in ascending order of price values ​​to obtain an ordered price signal sequence; The total weight of the signal records is obtained by summing the signal record weights of each price signal in the ordered price signal sequence. According to the order of the ordered price signal sequence, for each price signal, the signal record weights corresponding to the price signal and the price signals arranged before it are accumulated to obtain the cumulative weight corresponding to the price signal. Based on the ratio of the cumulative weight to the total weight of the signal records, the normalized cumulative weight corresponding to the price signal is determined. From the ordered price signal sequence, select the price signal whose normalized cumulative weight first reaches or exceeds the target weighted quantile, and determine the price value corresponding to the selected price signal as the potential customer's willingness to pay price.

[0010] In some optional implementations, the step of performing second quantile aggregation on the SKU willingness-to-pay price sample set at the SKU granularity to obtain the lower limit and upper limit of the guidance price range for the target SKU includes: Obtain the preset lower and upper quantiles; Arrange the potential customers' willingness to pay prices in the SKU willingness-to-pay price sample set in ascending order of price value to obtain an ordered willingness-to-pay price sequence; The lower limit of the guiding price range is determined from the ordered willingness-to-pay price sequence based on the lower quantile. The upper limit of the guidance price range is determined from the ordered willingness-to-pay price sequence based on the upper quantile.

[0011] In some optional implementations, the method further includes: Obtain a preset psychological price density construction method and density threshold, as well as potential customer feature information corresponding to each potential customer in the potential customer set, wherein the potential customer feature information includes at least one of behavior type, behavior timeliness and user profile; For each potential customer, the psychological price center and psychological price dispersion are determined based on the multi-source price signals corresponding to the potential customer. The psychological price center represents the price that the potential customer is most likely to accept, and the psychological price dispersion represents the degree of dispersion of the price that the potential customer can accept. Using the aforementioned psychological price density construction method, individual psychological price densities are established for each potential customer based on their corresponding psychological price center and psychological price dispersion. Based on the potential customer characteristic information corresponding to each potential customer, the corresponding potential customer weight is determined, and based on the potential customer weight corresponding to each potential customer, the individual psychological price density of each potential customer is weighted and aggregated to obtain the group consensus price density of the target SKU. The continuous price range in the consensus price density where the density value is not lower than the density threshold is determined as the theoretical pricing range of the target SKU.

[0012] In some optional implementations, the method further includes: The consensus price density of the group is accumulated along the direction of price increase to obtain the correspondence between price and cumulative probability. Obtain a preset set of candidate quantiles and a competition scoring method, as well as the market scarcity, same-store competition level, and exposure conversion efficiency of the target SKU; Using the aforementioned competition score determination method, the competition score corresponding to each candidate percentile in the candidate percentile set is determined based on the market scarcity, the same-store competition level, and the exposure conversion efficiency. The market scarcity and the exposure conversion efficiency have a positive impact on the competition score, while the same-store competition level has a negative impact on the competition score. Softmax processing is performed on the competitive scores corresponding to each candidate quantile, and the target quantile is determined from the candidate quantile set based on the processing results; Based on the correspondence between price and cumulative probability, the price corresponding to the cumulative probability reaching the target percentile is determined, and the price is determined as the adaptive recommended price of the target SKU.

[0013] Secondly, this application provides a vehicle price range prediction device, the device comprising: The data acquisition unit is used to acquire a target SKU, a set of potential customers corresponding to the target SKU, and a multi-source price signal corresponding to each potential customer in the set of potential customers, wherein the multi-source price signal includes price signals from at least two different sources. The first quantile aggregation unit is used to perform first quantile aggregation on the corresponding multi-source price signals at the user granularity for each potential customer, so as to obtain the potential customer's willingness to pay price. The sample set construction unit is used to construct a sample set of SKU payment willingness prices for the target SKU based on the potential customer's corresponding potential customer payment willingness price, wherein the sample set of SKU payment willingness prices includes at least one potential customer payment willingness price; The second quantile aggregation unit is used to perform second quantile aggregation on the SKU payment willingness price sample set at the SKU granularity to obtain the lower limit and upper limit of the guidance price range of the target SKU; The interval generation unit is used to generate the guidance price range for the target SKU based on the lower limit and the upper limit of the guidance price range.

[0014] In some optional implementations, the first quantile aggregation unit is specifically used for: Arrange the multi-source price signals corresponding to the potential customers in ascending order of price values ​​to obtain an ordered price signal sequence; Based on the preset lower truncation quantile and upper truncation quantile, the lower quantile threshold and upper quantile threshold of the ordered price signal sequence are determined respectively; Retain price signals in the ordered price signal sequence whose price values ​​are not lower than the lower quantile threshold and not higher than the upper quantile threshold; The median of the retained price signals is determined as the potential customer's willingness to pay price.

[0015] In some optional implementations, the first quantile aggregation unit is specifically used for: Obtain the signal source type and generation time of each price signal corresponding to the relevant potential customer, as well as the preset source credibility determination rules, exponential time decay method and target weighted quantile; The source reliability of each price signal is determined based on the signal source type and the source reliability determination rules. The time interval corresponding to each price signal is determined based on the generation time of each price signal and the current calculation time. Using the exponential time decay method, the source reliability of each price signal is time decayed according to the time interval corresponding to each price signal to obtain the signal recording weight of each price signal. Arrange the multi-source price signals corresponding to the potential customers in ascending order of price values ​​to obtain an ordered price signal sequence; The total weight of the signal records is obtained by summing the signal record weights of each price signal in the ordered price signal sequence. According to the order of the ordered price signal sequence, for each price signal, the signal record weights corresponding to the price signal and the price signals arranged before it are accumulated to obtain the cumulative weight corresponding to the price signal. Based on the ratio of the cumulative weight to the total weight of the signal records, the normalized cumulative weight corresponding to the price signal is determined. From the ordered price signal sequence, select the price signal whose normalized cumulative weight first reaches or exceeds the target weighted quantile, and determine the price value corresponding to the selected price signal as the potential customer's willingness to pay price.

[0016] In some optional implementations, the second quantile aggregation unit is specifically used for: Obtain the preset lower and upper quantiles; Arrange the potential customers' willingness to pay prices in the SKU willingness-to-pay price sample set in ascending order of price value to obtain an ordered willingness-to-pay price sequence; The lower limit of the guiding price range is determined from the ordered willingness-to-pay price sequence based on the lower quantile. The upper limit of the guidance price range is determined from the ordered willingness-to-pay price sequence based on the upper quantile.

[0017] In some optional embodiments, the device further includes a psychological price parameter determination unit, an individual psychological price density construction unit, a group consensus price density determination unit, and a theoretical pricing range determination unit.

[0018] The data acquisition unit is also used to acquire a preset psychological price density construction method and density threshold, as well as potential customer feature information corresponding to each potential customer in the potential customer set, wherein the potential customer feature information includes at least one of behavior type, behavior timeliness and user profile. The psychological price parameter determination unit is used to determine the psychological price center and psychological price dispersion for each potential customer based on the multi-source price signal corresponding to the potential customer. The psychological price center represents the price that the potential customer is most likely to accept, and the psychological price dispersion represents the degree of dispersion of the price that the potential customer can accept. The individual psychological price density construction unit is used to establish the individual psychological price density of each potential customer based on the psychological price center and psychological price dispersion of each potential customer using the psychological price density construction method. The group consensus price density determination unit is used to determine the corresponding potential customer weight based on the potential customer characteristic information corresponding to each potential customer, and to perform weighted aggregation of the individual psychological price density of each potential customer based on the potential customer weight corresponding to each potential customer, so as to obtain the group consensus price density of the target SKU. The theoretical pricing range determination unit is used to determine the continuous price range in the group consensus price density where the density value is not lower than the density threshold as the theoretical pricing range of the target SKU.

[0019] In some alternative implementations, the apparatus further includes an accumulation processing unit, a competitive scoring determination unit, a target quantile determination unit, and an adaptive recommendation price determination unit.

[0020] The cumulative processing unit is used to perform cumulative processing on the group consensus price density along the price increasing direction to obtain the correspondence between price and cumulative probability. The data acquisition unit is also used to acquire a preset set of candidate percentiles and a competitive scoring method, as well as the market scarcity, same-store competition level and exposure conversion efficiency of the target SKU; The competition score determination unit is used to determine the competition score corresponding to each candidate percentile in the candidate percentile set according to the competition score determination method, based on the market scarcity, the same-store competition level, and the exposure conversion efficiency. The market scarcity and the exposure conversion efficiency have a positive impact on the competition score, while the same-store competition level has a negative impact on the competition score. The target quantile determination unit is used to perform Softmax processing on the competitive scores corresponding to each candidate quantile, and determine the target quantile from the candidate quantile set based on the processing results; The adaptive recommendation price determination unit is used to determine the price corresponding to the cumulative probability reaching the target percentile based on the correspondence between price and cumulative probability, and to determine the price as the adaptive recommendation price of the target SKU.

[0021] Thirdly, this application provides an electronic device, comprising: One or more processors; Storage device, on which one or more programs are stored, When the above-mentioned one or more programs are executed by the above-mentioned one or more processors, the above-mentioned one or more processors implement the method as described in any embodiment of the first aspect of this application.

[0022] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by one or more processors, implements the method described in any embodiment of the first aspect of this application.

[0023] Fifthly, this application provides a computer program product, including a computer program / instructions, which, when executed by a processor, implement the method described in any embodiment of the first aspect of this application.

[0024] The vehicle price range prediction method and related products provided in this application first obtain the target SKU, the set of potential customers corresponding to the target SKU, and the multi-source price signals corresponding to each potential customer. Then, for each potential customer, the first quantile aggregation is performed on the corresponding multi-source price signals at the user granularity to obtain the potential customer's willingness to pay price, and a sample set of SKU willingness to pay prices for the target SKU is constructed based on the potential customer's willingness to pay prices. Next, the second quantile aggregation is performed on the SKU willingness to pay price sample set at the SKU granularity to obtain the lower limit and upper limit of the guidance price range for the target SKU. Finally, the guidance price range for the target SKU is generated based on the lower limit and upper limit of the guidance price range. This application, by obtaining price signals from different sources, can increase the data sources for vehicle price analysis and reduce the impact of insufficient data volume on the prediction results when using only a single source of price data. By first aggregating multi-source price signals corresponding to the same potential customer at the user granularity, each potential customer participates in subsequent processing with their corresponding willingness-to-pay price. This reduces data bias caused by duplicate inclusion of multiple price signals from the same potential customer, and makes the resulting willingness-to-pay price more fully represent the price acceptance of different potential customers for the target SKU. Further, by aggregating the willingness-to-pay prices of each potential customer at the SKU granularity, the impact of individual abnormal prices on the endpoints of the price range can be reduced, resulting in a more stable and reasonable range width for the generated guidance price range, thereby improving its actual reference value. Attached Figure Description

[0025] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings. The drawings are for illustrative purposes only and are not intended to limit the scope of this application. In the drawings: Figure 1 This is a schematic diagram of the system architecture in which the vehicle price range prediction method provided in this application embodiment can be applied; Figure 2 This is a flowchart illustrating the vehicle price range prediction method provided in the embodiments of this application; Figure 3 This is provided by the embodiments of this application. Figure 2 A schematic diagram of a decomposition process for step S202; Figure 4 This is provided by the embodiments of this application. Figure 2 A schematic diagram of a decomposition process for step S204; Figure 5 This is a schematic diagram of the vehicle price range prediction device provided in the embodiments of this application; Figure 6This is a schematic diagram of the structure of a computer system suitable for implementing the electronic devices of the present application embodiments. Detailed Implementation

[0026] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0027] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0028] Figure 1 An exemplary system architecture 100 is shown that can be applied to the vehicle price range prediction method provided in the embodiments of this application.

[0029] like Figure 1 As shown, system architecture 100 may include terminal devices 101, 102, and 103, network 104, and server 105. Network 104 is used to provide a communication link between terminal devices 101, 102, and 103 and server 105. Network 104 may include various communication connection types, such as wired communication links, wireless communication links, etc.

[0030] Users can use terminal devices 101, 102, and 103 to interact with server 105 via network 104 to receive or send data. Various client applications can be installed on terminal devices 101, 102, and 103, such as vehicle transaction applications, vehicle sales management applications, customer relationship management applications, vehicle inventory management applications, web browser applications, shopping applications, search applications, instant messaging tools, email clients, and social media platforms. When users perform actions such as clicking on, adding to favorites, leaving information, expressing interest, or making transactions through these client applications, price signals associated with the corresponding vehicle SKUs can be generated.

[0031] Terminal devices 101, 102, and 103 can be either hardware or software. When terminal devices 101, 102, and 103 are hardware, they can be various electronic devices with display and data interaction functions, including but not limited to smartphones, tablets, in-vehicle terminal devices, portable computers, and desktop computers. When terminal devices 101, 102, and 103 are software, they can be installed in the aforementioned electronic devices. Terminal devices 101, 102, and 103 can be implemented as multiple software programs or software modules, such as software modules for acquiring target SKUs, the potential customer set corresponding to the target SKUs, and multi-source price signals corresponding to each potential customer in the potential customer set. They can also be implemented as a single software program or software module; this application does not specifically limit their implementation in this regard.

[0032] Server 105 can be a server providing various data processing services, such as a backend server that processes the target SKU, the set of potential customers corresponding to the target SKU, and the multi-source price signals corresponding to each potential customer obtained or uploaded by terminal devices 101, 102, and 103. Server 105 can perform first quantile aggregation on the multi-source price signals corresponding to each potential customer at the user granularity to obtain the potential customer's willingness to pay price; construct a sample set of SKU willingness to pay prices for the target SKU based on the potential customer willingness to pay prices for each potential customer; perform second quantile aggregation on the sample set of SKU willingness to pay prices at the SKU granularity, and generate a guidance price range for the target SKU based on the obtained lower limit and upper limit of the guidance price range.

[0033] In some cases, the vehicle price range prediction method provided in this application can be jointly executed by terminal devices 101, 102, and 103 and server 105. For example, the steps of obtaining the target SKU, the potential customer set corresponding to the target SKU, and the multi-source price signals corresponding to each potential customer in the potential customer set can be executed by terminal devices 101, 102, and 103; the steps of performing first quantile aggregation at the user granularity, constructing a sample set of SKU payment willingness prices, performing second quantile aggregation at the SKU granularity, and generating a guiding price range can be executed by server 105. This application does not specifically limit this. Correspondingly, different functional modules in the vehicle price range prediction device can also be respectively set in terminal devices 101, 102, and 103 and server 105.

[0034] In some cases, the vehicle price range prediction method provided in this application can be executed by server 105, and correspondingly, the vehicle price range prediction device can also be set in server 105. In this case, system architecture 100 may not include terminal devices 101, 102, and 103, and server 105 can obtain the data required to execute the vehicle price range prediction method from other data acquisition devices or data storage systems.

[0035] In some cases, the vehicle price range prediction method provided in this application can be executed by terminal devices 101, 102, and 103, and correspondingly, the vehicle price range prediction device can also be set in terminal devices 101, 102, and 103. In this case, the system architecture 100 may not include server 105.

[0036] It should be noted that server 105 can be either hardware or software. When server 105 is hardware, it can be implemented as a distributed server cluster consisting of multiple servers, or as a single server. When server 105 is software, it can be implemented as multiple software programs or software modules, such as a software module for providing distributed data processing services, or as a single software program or software module. This application does not impose any specific limitations on this.

[0037] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown in the diagram is for illustrative purposes only. Depending on actual implementation needs, system architecture 100 may include any number of terminal devices, networks, and servers.

[0038] All information, data and signals involved in this application have been obtained with the user's authorization or with the full authorization of the relevant parties, and the collection, use and processing of the relevant data comply with the laws, regulations and standards of the relevant countries and regions.

[0039] Continue to refer to Figure 2 , Figure 2 The flowchart S200 of the vehicle price range prediction method provided in the embodiments of this application is shown.

[0040] Figure 2 The vehicle price range prediction method shown can be applied to Figure 1 The terminal device or server shown. This process S200 includes at least the following steps S201-S205.

[0041] S201, obtain the target SKU, the set of potential customers corresponding to the target SKU, and the multi-source price signals corresponding to each potential customer in the set of potential customers.

[0042] In this embodiment, the target SKU is the vehicle SKU for which the current price range for which guidance needs to be predicted. A vehicle SKU can be used to identify a vehicle object with a specific combination of vehicle attributes. These attributes can include one or more of the following: vehicle brand, model series, vehicle type, model year, and configuration. Different vehicle SKUs can correspond to different combinations of vehicle attributes.

[0043] Potential customers refer to users who have engaged in activities related to the target SKU. These activities may include clicks, favorites, contact information, expressions of interest, vehicle transactions, or other behaviors that indicate a user's interest in the target SKU. The set of potential customers corresponding to the target SKU can be determined based on users who have engaged in activities related to the target SKU within a preset statistical time frame, and the set must include at least one potential customer.

[0044] Specifically, terminal devices or servers can query data records associated with the target SKU from the vehicle trading platform's behavior logs, transaction records, or customer management data based on the SKU identifier of the target SKU. User identifiers are extracted from the retrieved data records, and potential customers corresponding to the target SKU are identified based on these user identifiers, thus forming a set of potential customers for the target SKU.

[0045] Multi-source price signals include price signals from at least two different sources. Multi-source price signals refer to price data associated with potential customers and target SKUs obtained from at least two different sources. The sources of price signals can include at least two of the following: transaction price, click price, favorites price, lead generation price, and interest price.

[0046] The transaction price can be the actual transaction price formed when a potential customer purchases a vehicle corresponding to the target SKU.

[0047] The click price can be the displayed price of a vehicle when a potential customer clicks on the target SKU.

[0048] The collection price can be the price corresponding to the vehicle when a potential customer collects the target SKU.

[0049] The lead generation price can be the price that a potential customer pays for a vehicle when they submit their contact or inquiry information for a target SKU.

[0050] Interest pricing can be the vehicle price associated with a potential customer's expression of interest in a target SKU.

[0051] In some embodiments, chat logs, phone transcribed texts, or contact information between potential customers and customer service representatives or dealers on the vehicle trading platform can also be obtained, and a price signal extraction prompt template can be constructed. The price signal extraction prompt template can be used to instruct a large language model to identify at least one structured field from unstructured content, including target price, psychological price ceiling, psychological price floor, and budget constraint.

[0052] The structured fields output by large language models can be mapped to chat-based price signals. Appropriate signal source types and source credibility can be configured for these chat-based price signals, and they can be added to the multi-source price signals corresponding to potential customers. For outputs containing psychological price upper and lower limits, the psychological price upper and lower limits, or the price center determined by these two limits, can be used as price signals respectively.

[0053] Chat consultation price signals can be processed together with transaction price, click price, favorite price, lead generation price, and interest price signals to perform data preprocessing, sorting, quantile filtering, and potential customer willingness to pay price determination processes; when using weighted quantile aggregation, the corresponding signal record weights can also be determined based on the source credibility and the time of generation of the chat consultation content.

[0054] Each price signal can correspond to a user identifier, SKU identifier, price value, signal source type, and generation time.

[0055] Terminal devices or servers can associate each price signal with the corresponding potential customer and target SKU based on the user identifier and SKU identifier, thereby obtaining the multi-source price signal corresponding to each potential customer in the potential customer set.

[0056] A single potential customer can correspond to one or more price signals, and the multi-source price signals corresponding to a particular potential customer include price signals from at least two different sources.

[0057] In some embodiments, after acquiring the multi-source price signals corresponding to each potential customer, the acquired data can be preprocessed. Preprocessing may include deleting data records lacking user identifiers, SKU identifiers, or price values; standardizing the pricing units, currencies, or data formats of price signals from different sources; and normalizing user identifiers and SKU identifiers. The preprocessed multi-source price signals can then be used as input for subsequent first quantile aggregation.

[0058] By acquiring price signals from different sources between potential customers and target SKUs, we can increase the data sources for analyzing potential customer price acceptance and reduce the impact of insufficient data volume when using only historical transaction data. Furthermore, establishing the correspondence between potential customers, target SKUs, and price signals through user and SKU identifiers provides a data foundation for subsequent processing of multi-source price signals for each potential customer at the user granularity.

[0059] S202, for each potential customer, perform first quantile aggregation on the corresponding multi-source price signals at the user granularity to obtain the potential customer's willingness to pay price.

[0060] In this embodiment, the first quantile aggregation is used to convert multiple price signals generated by the same potential customer for a target SKU into a single potential customer's willingness to pay price.

[0061] The first quantile aggregation is performed independently for each potential customer in the potential customer set, and the price signals corresponding to different potential customers are not mixed in this step.

[0062] For potential customers and target SKU The multi-source price signals generated by the potential customer for the target SKU can be represented as: in, Indicates potential customer Corresponding multi-source price signals; Price signals indicating the source of the transaction price; This indicates the price signal from which the click price originates; Price signals indicating the source of a collectible's value; Price signals indicating the source of retained earnings; Price signals that indicate the source of interest prices.

[0063] The same source can include one or more price signals. For example, when a potential customer clicks on the vehicle corresponding to the target SKU multiple times, multiple price signals from different click price sources can be generated.

[0064] refer to Figure 3 , Figure 3 This is provided by the embodiments of this application. Figure 2 A schematic diagram of a decomposition process for step S202.

[0065] In one implementation, such as Figure 3 As shown, step S202 may include the following steps S2021-S2024.

[0066] S2021, arrange the multi-source price signals corresponding to the potential customers in ascending order of price values ​​to obtain an ordered price signal sequence.

[0067] In this embodiment, the preprocessed multi-source price signals can be arranged in ascending order of price value. Let the number of valid price signals be... The resulting ordered price signal sequence can be represented as: in, Represents the first in an ordered price signal sequence The price value corresponding to each price signal.

[0068] S2022, Based on the preset lower truncation quantile and upper truncation quantile, determine the lower quantile threshold and upper quantile threshold of the ordered price signal sequence, respectively.

[0069] In this application's embodiments, quantile points and quantile thresholds are parameters of different types. A quantile point represents the proportional position of a price signal within an ordered price signal sequence and is a dimensionless proportional parameter. A quantile threshold is a price value determined from the ordered price signal sequence based on the corresponding quantile point, and its unit of measurement is the same as that of the price signal.

[0070] Specifically, obtain the preset lower cutoff quantile. and upper cutoff quantile ,in, .

[0071] Lower cutoff quantile Upper truncation quantile is used to indicate the cutoff position near the low price end in an ordered price signal sequence. Used to indicate the cutoff position near the high price end in an ordered price signal sequence.

[0072] For ordered price signal sequences Perform and lower cutoff quantiles respectively and upper cutoff quantile The corresponding quantile determination operation yields the lower quantile threshold. and upper quantile threshold lower quantile threshold The upper quantile threshold is used to define the lower price boundary of the price signal to be retained. Used to define the upper price boundary of the price signal to be retained.

[0073] Lower quantile threshold and upper quantile threshold It can be represented as: in, Indicates the potential customers currently being processed; Indicates potential customers The corresponding multi-source price signals are arranged in ascending order of price values ​​to obtain an ordered price signal sequence; Indicates the sequence of ordered price signals Determine the lower cutoff quantile The operation of determining the quantiles of the corresponding price values; Indicates the sequence of ordered price signals Determine the upper cutoff quantile. The operation of determining the quantile of the corresponding price value.

[0074] In one alternative quantile determination rule, linear interpolation can be used to determine the quantile threshold. This applies to an ordered price signal sequence comprising m price values ​​and any quantile point. First, the quantile position is determined according to (m-1)×q+1. When the quantile position is an integer, the price value corresponding to that position is determined as the quantile threshold; when the quantile position is not an integer, the quantile threshold is obtained by linear interpolation based on the two adjacent price values ​​on both sides of the quantile position and the decimal part of the quantile position. The lower quantile threshold and the upper quantile threshold in the above example are both determined according to this linear interpolation method.

[0075] In one example, the lower cutoff quantile It can be 25%, upper cutoff quantile. It can be 75%. In other examples, the lower cutoff quantile and the upper cutoff quantile can also be 10% and 90%, respectively, depending on the quantity and distribution of price signals.

[0076] S2023, retain price signals in the ordered price signal sequence whose price values ​​are not lower than the lower quantile threshold and not higher than the upper quantile threshold.

[0077] Specifically, price signals whose price values ​​fall between the lower and upper quantile thresholds can be retained, and the retained price signals can form a set of filtered price signals. .

[0078] Filtered price signal set It can be represented as: Through the above processing, price signals with prices below the lower quantile threshold or above the upper quantile threshold can be filtered out, thereby reducing the impact of extreme price signals at both ends of the ordered price signal sequence on the determination of potential customers' willingness to pay.

[0079] S2024, the median of the retained price signals is determined as the potential customer's willingness to pay price.

[0080] The corresponding potential customer's willingness to pay price can be expressed as: in, Indicates potential customer Target SKU The price that potential customers are willing to pay. This represents the set of filtered price signals. The price value obtained by performing the median determination operation on the price value in the middle.

[0081] For example, suppose a potential customer generates multi-source price signals for a target SKU, including a click price of 135,000 yuan, a collection price of 140,000 yuan, a click price of 145,000 yuan, a lead generation price of 148,000 yuan, a click price of 150,000 yuan, an interest price of 152,000 yuan, and a noise price of 250,000 yuan. Arranged in ascending order of price values, this yields an ordered price signal sequence: ={13.5,14.0,14.5,14.8,15.0,15.2,25.0} With the lower quantile threshold set to 25% and the upper quantile threshold set to 75%, the determined lower quantile threshold is approximately 142,500 yuan, and the upper quantile threshold is approximately 151,000 yuan. After quantile filtering, a set of filtered price signals is obtained. {14.5,14.8,15.0}.

[0082] Therefore, the potential customer's willingness to pay for the target SKU is 148,000 yuan.

[0083] In the example above, if the average of all price signals before filtering is taken directly, the result is 160,000 yuan, which is easily affected by the higher price signal of 250,000 yuan. By first performing quantile filtering and then taking the median of the retained price signals, the influence of price signals at both ends of the price signal sequence on the aggregation result can be reduced, so that the determined potential customer's willingness to pay price can better reflect the corresponding potential customer's price acceptance level.

[0084] In another implementation, the corresponding signal recording weight can be determined for each price signal based on the signal source type and generation time of each price signal, and the potential customer's willingness to pay price can be determined based on the signal recording weight of each price signal.

[0085] Specifically, the signal source type and generation time of each price signal corresponding to the potential customer are obtained, as well as the preset source credibility determination rules, exponential time decay method and target weighted quantile.

[0086] Here, the source credibility determination rule can record the correspondence between different signal source types and their corresponding source credibility; the target weighted quantile can be a pre-set value between 0 and 1.

[0087] The source reliability of each price signal is determined based on the signal source type and source reliability determination rules.

[0088] For example, different signal source types can correspond to different source credibility, and the source credibility can decrease in the following order: transaction price source, comment price source, favorite price source, click price source, and interest price source.

[0089] The time interval corresponding to each price signal is determined based on the generation time of each price signal and the current calculation time.

[0090] In this embodiment of the application, for the first For each price signal, the time difference between the time the price signal is generated and the current calculation time can be defined as the corresponding time interval. .

[0091] The exponential time decay method is used to perform time decay processing on the source reliability of each price signal according to the time interval corresponding to each price signal, so as to obtain the signal record weight of each price signal.

[0092] No. The signal record weights of each price signal can be represented as: in, Indicates the first The signal record weights for each price signal; Indicates according to the first The reliability of a price signal source is determined by the type of signal source it originates from; Indicates the first The time interval between the generation time of a price signal and the current calculation time; Indicates the time decay coefficient; This represents the exponential time decay factor.

[0093] As the time interval increases, the exponential time decay factor decreases, resulting in earlier price signals receiving lower signal recording weights.

[0094] Arrange the multi-source price signals corresponding to the potential customers in ascending order of price values ​​to obtain an ordered price signal sequence.

[0095] After sorting, the signal source type, generation time, and signal record weight of each price signal are adjusted and arranged in the same order as the corresponding price signal.

[0096] The total weight of the signal records is obtained by summing the signal record weights of each price signal in the ordered price signal sequence.

[0097] The total weight of the signal recording can be expressed as: in, This indicates the total weight of the signal record; Indicates the order in the order of The weight of the signal record corresponding to the price signal of a bit; This represents the position index of the price signal during the summation process.

[0098] According to the order of the ordered price signal sequence, for each price signal, the signal record weights corresponding to the price signal and the price signals preceding it are accumulated to obtain the cumulative weight of the price signal. Based on the ratio of the cumulative weight to the total weight of the signal records, the normalized cumulative weight of the price signal is determined.

[0099] No. The cumulative weight and normalized cumulative weight corresponding to each price signal can be expressed as follows: in, Indicates the first The cumulative weight of each price signal; Indicates the first The normalized cumulative weights corresponding to each price signal.

[0100] The normalized cumulative weights of each price signal increase sequentially in the direction of increasing price values.

[0101] From the ordered price signal sequence, select the price signal whose normalized cumulative weight first reaches or exceeds the target weighted quantile, and determine the price value corresponding to the selected price signal as the potential customer's willingness to pay price.

[0102] The corresponding potential customer's willingness to pay price can be expressed as: in, This indicates the price value corresponding to the selected price signal; This indicates the position of the selected price signal within the ordered price signal sequence; This represents the target weighted quantile. When the target weighted quantile is 0.5, the determined potential customer's willingness to pay price can be regarded as the weighted median price of the corresponding potential customer.

[0103] By using the weighted quantile aggregation method described above, price signals with higher credibility and more recent generation time can have a greater impact on potential customers' willingness to pay, while reducing the impact of price signals with lower credibility or earlier generation time on the aggregation results.

[0104] The aforementioned weighted quantile aggregation can be implemented using either of the following two methods. In the first method, the pre-quantile filtering shown in steps S2022 and S2023 can be omitted. All valid multi-source price signals corresponding to the relevant potential customer can be directly used as the object of weighted quantile processing, and the signal record weight of each price signal can be determined based on the signal source type and generation time. In the second method, steps S2021 to S2023 can be executed first to obtain a set of filtered price signals. Then, the signal record weight of each price signal in the filtered price signal set can be determined, and weighted quantile processing can be performed. The result of the weighted quantile processing replaces the median in step S2024. Both methods are used to determine the potential customer's willingness to pay price at the user granularity, and can be selected based on the number of price signals, the frequency of abnormal price occurrences, or business processing requirements.

[0105] In some embodiments, the signal record weights of different price signals can be set as learnable parameters, and the weight network can be trained using historical vehicle transaction data. Specifically, training samples can be constructed, using the actual transaction price of the target SKU as the supervision signal and the multi-source price signal tensor generated by each potential customer for the target SKU as the input. The multi-source price signal tensor may include the price value of the price signal, the signal source type encoding, the time interval of the price signal from the current calculation time, and the user profile information of the corresponding potential customer.

[0106] The weighted network can include a multilayer perceptron (MLP) and a normalization layer. For each price signal, the signal source type is encoded. Time interval and user profile The input is a multilayer perceptron, and the output of the multilayer perceptron is processed using Softmax to obtain the normalized signal recording weights corresponding to each price signal. This process can be expressed as: Here, w represents a weight vector composed of the signal record weights corresponding to each price signal. During training, the signal record weights output by the weight network can be used to determine the potential customer's willingness to pay price for each potential customer according to the aforementioned weighted quantile processing procedure. The potential customer's willingness to pay price is then input into the second quantile aggregation process at the SKU granularity. Therefore, the weight network is used to determine the signal record weights in the first quantile aggregation, rather than directly weighting and summing multiple price signals to obtain the source fusion center price.

[0107] A joint loss can be constructed based on the interval width loss and interval miss loss of the SKU-level price range, and the weight network parameters can be updated using the joint loss. The joint loss can be expressed as: in, This represents the interval width loss used to constrain the width of the guiding price range; This indicates a loss incurred when the actual transaction price does not fall within the guidance price range; and These represent the weight coefficients of the corresponding loss terms. During online processing, the trained weighted network can be used to determine the signal record weights of the real-time price signal, and then weighted quantile processing can be performed.

[0108] In some embodiments, a minimum number of valid signals can also be set. When the number of price signals retained after quantile filtering is less than the minimum number of valid signals, the corresponding potential customer can be marked as an invalid potential customer, and the potential customer's willingness to pay price will not be used for subsequent second quantile aggregation at the SKU granularity. The minimum number of valid signals can be, for example, 2, or can be set according to the actual data scale.

[0109] By processing the multi-source price signals corresponding to each potential customer at the user level, multiple price signals generated by the same potential customer can be converted into a single potential customer's willingness to pay price, avoiding excessive influence from a large number of potential customers' price signals on subsequent processing. Employing quantile filtering or weighted quantile processing can further reduce the impact of extreme prices, low-reliability price signals, and earlier-generated price signals on the potential customer's willingness to pay price.

[0110] S203, construct a sample set of SKU payment willingness prices for the target SKU based on the payment willingness price of each potential customer.

[0111] The SKU payment willingness price sample set includes at least one potential customer payment willingness price.

[0112] Specifically, after determining the potential customer's willingness to pay price for each potential customer in step S202, the potential customer's willingness to pay price for each potential customer can be aggregated according to the target SKU.

[0113] Target SKU Obtain the target SKU Corresponding potential customers The potential customers' willingness to pay price is obtained, and the obtained potential customers' willingness to pay price constitutes a sample set of SKU willingness to pay prices. .

[0114] SKU Willingness to Pay Price Sample Set It can be represented as: in, Indicates the target SKU The corresponding SKU payment willingness price sample set; Indicates the target SKU The corresponding number A potential customer; Indicates potential customer ᵢTargeting the target SKU The price that potential customers are willing to pay; This represents the number of potential customers' willingness to pay prices in the SKU willingness-to-pay price sample set, and It is an integer greater than or equal to 1.

[0115] The multi-source price signals corresponding to each potential customer are aggregated at the first quantile level to form a potential customer's willingness to pay price. Therefore, when constructing the SKU willingness to pay price sample set, the willingness to pay price of each potential customer is added to the sample set as an independent sample. When different potential customers have the same willingness to pay price, the willingness to pay prices of each potential customer are still retained as independent samples to reflect the frequency of occurrence of the same price acceptance level in the potential customer set.

[0116] For example, suppose the target SKU For three potential customers, their willingness to pay is 151,000 yuan, 148,000 yuan, and 135,000 yuan respectively. Therefore, the target SKU... The corresponding SKU willingness-to-pay price sample set can be represented as: It should be noted that in step S203, only the potential customer's willingness to pay price is aggregated to construct a sample set of SKU willingness to pay prices; the order of the potential customer's willingness to pay prices in this sample set is not limited. Sorting the SKU willingness to pay price sample set and determining the corresponding quantile positions can be performed in the subsequent second quantile aggregation process at the SKU granularity.

[0117] By aggregating the willingness-to-pay prices of each potential customer into the same sample set corresponding to the target SKU, each potential customer can participate in subsequent processing with a single willingness-to-pay price. This avoids the excessive influence of potential customers with a larger number of price signals on the SKU granular aggregation results due to the different numbers of original price signals generated by different potential customers, and provides a sample basis for subsequently determining the lower limit and upper limit of the guidance price range for the target SKU.

[0118] S204, at the SKU granularity, performs second quantile aggregation on the sample set of SKU payment willingness prices to obtain the lower limit and upper limit of the guidance price range for the target SKU.

[0119] It should be noted that the first quantile aggregation in step S202 uses the multi-source price signal corresponding to a single potential customer as the processing object, and is used to determine the payment willingness price of a potential customer. The second quantile aggregation in step S204 uses the SKU payment willingness price sample set constructed in step S203 as the processing object, and is used to determine the two endpoints of the guidance price range among multiple potential customers corresponding to the target SKU.

[0120] The second quantile aggregation means that after the first quantile aggregation at the user granularity, the payment willingness price of multiple potential customers corresponding to the target SKU is processed again by quantile.

[0121] refer to Figure 4 , Figure 4 This is provided by the embodiments of this application. Figure 2 A schematic diagram of a decomposition process for step S204.

[0122] Step S204 may include the following steps S2041 to S2044.

[0123] S2041, obtain the preset lower quantile and upper quantile.

[0124] In this embodiment, the lower quantile represents the proportion of the ordered willingness-to-pay price sequence closer to the low price end, and the upper quantile represents the proportion of the ordered willingness-to-pay price sequence closer to the high price end; both are dimensionless proportions and not specific prices.

[0125] The lower quantile can be represented as The upper quantile can be represented as Both the lower and upper quantiles can be preset values ​​between 0 and 1, with the lower quantile being smaller than the upper quantile.

[0126] For example, the lower quantile It can be set to 10%, upper quantile. It can be set to 90%. In other implementations, the lower and upper quantiles can be adjusted between 5% and 95% to 20% and 80%, depending on the requirements for price coverage and price range width. For example, the lower and upper quantiles can be set to 5% and 95%, 15% and 85%, or 20% and 80%, respectively; there are no restrictions here.

[0127] In this design, a smaller lower quantile and a larger upper quantile allow the two interval endpoints to cover more potential customers' willingness to pay; conversely, a larger lower quantile and a smaller upper quantile can reduce the distance between the two interval endpoints. Therefore, appropriate quantiles can be set according to the coverage and interval width requirements in practical applications.

[0128] S2042, sort the potential customers' willingness to pay prices in the SKU willingness to pay price sample set in ascending order of price value to obtain an ordered willingness to pay price sequence.

[0129] In this embodiment of the application, the sorting only changes the position of each potential customer's willingness to pay price in the sequence, without changing the potential customer's willingness to pay price itself, and without merging the same price value corresponding to different potential customers.

[0130] Here, the ordered price series of willingness to pay can be represented as .

[0131] When determining the endpoints of the second quantile aggregation interval using the above linear interpolation method, the number of potential customers' willingness to pay prices in the SKU willingness to pay price sample set can be used as the sequence length of the ordered willingness to pay price sequence. The lower quantile and the upper quantile are then substituted into the above quantile position determination process to obtain the lower limit and upper limit of the guidance price interval.

[0132] S2043, based on the lower quantile, determines the lower limit of the guidance price range from the ordered willingness-to-pay price sequence.

[0133] In this embodiment of the application, the lower limit of the guidance price range represents the specific price value corresponding to the lower percentile.

[0134] For example, when the lower quantile is 10%, it means that the price value corresponding to the 10th percentile position at the lower end of the ordered willingness-to-pay price sequence needs to be determined.

[0135] Based on the lower quantile, starting from the low end of the ordered willingness-to-pay price sequence, determine the quantile position corresponding to the lower quantile, and obtain the price value corresponding to that quantile position according to the preset quantile determination rules. When the quantile position corresponds to a potential customer's willingness-to-pay price in the ordered willingness-to-pay price sequence, that potential customer's willingness-to-pay price can be determined as the lower limit of the guidance price range; when the quantile position is between two adjacent potential customer willingness-to-pay prices, the corresponding price value can be determined based on the two adjacent potential customer willingness-to-pay prices according to the preset quantile determination rules.

[0136] The lower limit of the guidance price range can be expressed as: in, Indicates the target SKU The lower limit of the guidance price range; Indicates the lower quantile; Indicates the price series of ordered willingness to pay. The determined quantiles at the lower end The corresponding price value. Therefore, the impact of the willingness-to-pay prices of a small number of potential customers located at the lower end of the ordered willingness-to-pay price sequence on the lower limit of the guidance price range can be reduced.

[0137] S2044, based on the upper quantile, determines the upper limit of the guidance price range from the ordered willingness-to-pay price sequence.

[0138] The upper limit of the guidance price range represents the specific price value corresponding to the upper percentile. The units of measurement for both the upper and lower limits of the guidance price range are the same as those for the potential customer's willingness to pay.

[0139] For example, when the upper quantile is 90%, it means that the price value corresponding to the 90th quantile position needs to be determined from the ordered willingness-to-pay price sequence.

[0140] Based on the upper quantile, the corresponding quantile position in the ordered willingness-to-pay price sequence is determined, and the price value corresponding to that quantile position is obtained according to a preset quantile determination rule. When the quantile position corresponds to a potential customer's willingness-to-pay price in the ordered willingness-to-pay price sequence, that potential customer's willingness-to-pay price can be determined as the upper limit of the guidance price range; when the quantile position is between two adjacent potential customer willingness-to-pay prices, the corresponding price value can be determined based on the two adjacent potential customer willingness-to-pay prices according to the preset quantile determination rule.

[0141] The upper limit of the guidance price range can be expressed as: in, Indicates the target SKU The upper limit of the guidance price range; Indicates the upper quantile; Indicates the price series of ordered willingness to pay. The determined quantiles at the upper end The corresponding price value. Therefore, the impact of the willingness-to-pay prices of a small number of potential customers located at the high end of the ordered willingness-to-pay price sequence on the upper limit of the guidance price range can be reduced.

[0142] For example, suppose the target SKU The corresponding SKU willingness-to-pay price sample set is as follows: After sorting the prices in ascending order, we obtain an ordered price sequence of willingness to pay: In this small sample scenario, the lower quantile can be set to 0, and the upper quantile to 1, that is, the minimum and maximum price values ​​in the ordered willingness-to-pay price sequence can be selected as the two interval endpoints. Therefore, it can be determined that: When the SKU payment willingness price sample set includes only one potential customer payment willingness price, the price values ​​corresponding to the lower percentile and the upper percentile can be the same, and the lower limit and upper limit of the determined guidance price range can also be the same.

[0143] By aggregating the willingness-to-pay prices of potential customers at the SKU level using the second quantile, the impact of the willingness-to-pay prices of a small number of low-end or high-end potential customers on the endpoints of the range can be reduced. This ensures that the determined lower and upper limits of the guidance price range reflect the distribution of the willingness-to-pay prices of potential customers corresponding to the target SKU, providing the endpoints for generating the guidance price range for the target SKU in the future.

[0144] In some embodiments, the potential customer set can be divided into K profile clusters based on at least one user profile tag selected from income level, city tier, historical transaction price range, and activity level. to Each profile cluster is assigned a priori weight. The closer the price positioning of a profile cluster matches the target SKU, the greater the priori weight of the corresponding profile cluster can be.

[0145] In the second quantile aggregation process at the SKU granularity, the SKU willingness-to-pay price sample set can be expanded into a weighted willingness-to-pay price sample set: in, Indicates potential customer The prior weights corresponding to the respective customer profile clusters. Based on the willingness-to-pay price of each potential customer and its corresponding prior weights, a weighted empirical cumulative distribution of the target SKU can be established: in, Indicates the target SKU in terms of price The weighted cumulative probability of experience at the location; Indicates potential customer The corresponding prior weights; here, and Indicates the same weight, i.e. = The parentheses {·} represent an indicative operation, taking a value of 1 if the condition within the parentheses is true, and 0 otherwise. This can be used to indicate when the weighted empirical cumulative probability first reaches or exceeds the lower quantile. and upper quantile The corresponding prices are determined as the lower and upper limits of the guide price range. This implementation method introduces user profile layering weights in the second stage, enabling the guide price range of the target SKU to reflect the matching relationship between different target groups and the vehicle's value positioning.

[0146] S205, Generate the guidance price range for the target SKU based on the lower limit and upper limit of the guidance price range.

[0147] The guidance price range represents a continuous price range defined by the lower limit and the upper limit of the guidance price range. It is used to characterize the price range in which the willingness of potential customers to pay for the target SKU is relatively concentrated.

[0148] When the lower limit of the guidance price range is not higher than the upper limit, the lower limit can be used as the left endpoint of the range, and the upper limit as the right endpoint, to generate the guidance price range for the target SKU. The guidance price range can be represented as: in, Indicates the target SKU The guidance price range.

[0149] For example, for a specific target SKU, if the second quantile aggregation at the SKU granularity determines a lower limit of the guidance price range as RMB 135,000 and an upper limit of RMB 151,000, then the guidance price range for that target SKU can be generated as [RMB 135,000, RMB 151,000]. This guidance price range indicates that, based on the potential customer willingness-to-pay price sample corresponding to the target SKU, RMB 135,000 to RMB 151,000 can serve as the guidance price range for the target SKU.

[0150] In some embodiments, after generating the guide price range, the identification information of the target SKU, the lower limit of the guide price range, the upper limit of the guide price range, and the guide price range can be associated, and the association result can be stored in a preset database. The identification information of the target SKU may include at least one of the following: vehicle model identifier, configuration version identifier, region identifier, dealer identifier, or vehicle inventory unit identifier.

[0151] In some embodiments, the guiding price range can also be sent to the terminal device and displayed through the terminal device's display interface. The display format may include at least one of the following: range text, price range bar, or price distribution chart.

[0152] For example, the "Guidance Price Range: RMB 135,000 to RMB 151,000" can be displayed on the vehicle information page corresponding to the target SKU, so that relevant personnel can conduct price analysis or make pricing decisions based on this guidance price range.

[0153] It should be noted that the guidance price range generated in step S205 comes from the first quantile aggregation at the user granularity and the second quantile aggregation at the SKU granularity. This is the output of a different processing path than the theoretical pricing range determined based on the group consensus price density. The two can be stored, displayed, or applied separately.

[0154] By generating a guidance price range for the target SKU based on the lower and upper limits of the guidance price range, multi-source price signals from different potential customers can be converted into a price range with clear boundaries. This reduces the impact of a single abnormal price signal or a large number of repetitive price signals generated by some potential customers on the range result, enabling the generated guidance price range to reflect the price acceptance of the target SKU by multiple potential customers more stably.

[0155] In some embodiments, vehicle price range prediction can also be modeled as a machine learning task that directly predicts the endpoints of the range. This implementation is an alternative prediction path parallel to the two-stage quantile aggregation path shown in steps S201 to S205, and is not a processing step in either the first quantile aggregation or the second quantile aggregation.

[0156] It can obtain at least one structured vehicle feature of the target SKU, including the vehicle series, brand, model, condition, mileage, and year, as well as the mean, standard deviation, and at least one statistical feature of the potential customer's willingness to pay price distribution obtained by first quantile aggregation, and the above features constitute the model input.

[0157] Training can be performed separately on the lower quantiles. and upper quantile The corresponding lower quantile regression model and upper quantile regression model. For any target quantile... The corresponding quantile regression model can be trained by minimizing the quantile loss function, which can be expressed as: in, This represents the actual transaction price corresponding to the training sample; This represents the predicted price output by the quantile regression model; This represents the target quantile for the corresponding model. During online processing, the input features corresponding to the target SKU can be input into the lower quantile regression model and the upper quantile regression model respectively to obtain the lower predicted price and the upper predicted price. The lower predicted price and the upper predicted price are then determined as the lower limit and upper limit of the model's predicted price range, respectively.

[0158] The quantile loss function is used to assign different penalty weights to two types of deviations based on the target quantile: predicted price being lower than the actual transaction price and predicted price being higher than the actual transaction price. The higher the target quantile, the greater the penalty for a lower predicted price, and the lower the target quantile, the greater the penalty for a higher predicted price.

[0159] The guidance price range obtained from steps S201 to S205 can serve as a fallback result or monitoring baseline for the quantile regression prediction path. For example, when the model input is missing, the lower limit of the model output range is higher than the upper limit, or the model output exceeds the preset effective business price range, the guidance price range obtained from the two-stage quantile aggregation can be output; alternatively, the deviation between the endpoints of the two paths can be compared, and a data quality alert can be generated when the deviation exceeds a preset threshold.

[0160] In some embodiments, the price signals corresponding to the target SKU can be grouped according to the signal source type of the price signals to obtain price signal groups corresponding to each price signal source.

[0161] For each price signal source, the price signals in the corresponding price signal group can be aggregated to obtain the source-level price. Aggregation methods can include median aggregation, quantile aggregation, or weighted quantile aggregation.

[0162] Furthermore, the corresponding source weights can be determined based on the reliability, timeliness, and relevance of each price signal source. These source weights are then normalized to ensure they are non-negative and that the sum of all source weights is one. Specifically, reliability characterizes the reliability of the price information provided by the corresponding price signal source, timeliness characterizes the relevance of the price information to the current calculation time, and relevance characterizes the degree of association between the corresponding price signal source and the target SKU.

[0163] The source fusion center price of the target SKU is obtained by weighting and fusing the primary prices and weights of each price signal source.

[0164] When the number of potential customers corresponding to a target SKU is lower than the preset minimum number of potential customers, a safety net price range for the target SKU can be determined based on the source fusion center price and the market breadth prior of the vehicle series to which the target SKU belongs. The market breadth prior characterizes the range of price fluctuations that the vehicle series to which the target SKU belongs typically experiences around the center price.

[0165] In one example, an effective potential customer can be one who has not been marked as an invalid potential customer and whose willingness to pay has been confirmed. The minimum number of effective potential customers can be set to 5. The center of the price range for each vehicle within the target SKU's vehicle series can be determined based on their price ranges. The average of the center price range and the source fusion center price is used as the safety net center price. Based on a market width prior of 10% to 15% fluctuation between the vehicle series' market price and the center price, the lower and upper limits of the safety net guidance price range are determined.

[0166] The source fusion center price can also serve as a verification price for the two-stage quantile aggregation results. Specifically, the median of the ordered willingness-to-pay price series can be used to determine the reference value of the central interval for the guiding price range, and the relative deviation between the source fusion center price and the reference value can be compared. When the relative deviation exceeds a preset deviation threshold, a data quality alert can be generated or a manual review can be triggered. The preset deviation threshold can be, for example, 20%.

[0167] The source dimension fusion path can serve as an auxiliary path for the two-stage quantile aggregation path or a fallback path for small samples, without altering the process of determining the guiding price range formed in steps S201 to S205.

[0168] In some embodiments, the vehicle price range prediction method can also determine the theoretical pricing range for the target SKU based on the psychological price distribution of each potential customer. This processing path may include the following steps A1 to A5.

[0169] It should be noted that the theoretical pricing range formed by A1 to A5 is based on the price density of group consensus, which belongs to a different processing path than the guiding price range formed by two-stage quantile aggregation in steps S201 to S205. The endpoints of the theoretical pricing range and the guiding price range can be different, and they can be output separately or used in different price analysis scenarios.

[0170] A1, obtain the preset psychological price density construction method and density threshold, as well as the potential customer characteristic information corresponding to each potential customer in the potential customer set.

[0171] Among them, potential customer characteristic information includes at least one of behavior type, behavior timeliness, and user profile.

[0172] The psychological price density construction method represents a method for establishing a correspondence between a price and the relative likelihood of a potential customer accepting that price, based on the potential customer's psychological price center and psychological price dispersion.

[0173] In one implementation, the psychological price density can be constructed using a Gaussian distribution.

[0174] The density threshold represents the density height criterion used to identify high-density price ranges from the consensus price density. The density threshold can be set to a preset fixed density value, or it can be set based on the maximum density value of the consensus price density.

[0175] For example, the density height corresponding to 80% of the maximum density value can be set as the density threshold. Here, 80% represents the proportion relative to the maximum density height, not the cumulative probability of reaching 80%.

[0176] Potential customer characteristics are used to characterize the behavior of potential customers when they generate price signals and the degree of association between potential customers and target SKUs.

[0177] Among them, the types of behavior can include clicks, favorites, leaving information, transactions, or other behaviors that can reflect the intention to purchase a vehicle.

[0178] Behavioral timeliness can characterize the degree of temporal proximity between the moment when the corresponding behavior occurs and the current calculation moment.

[0179] User profiles can include at least one of the following: income level, city tier, historical transaction price range, and activity level.

[0180] In some embodiments, one of the following can be obtained as potential customer characteristic information: behavior type, behavior timeliness, and user profile; or two or three of these information can be used in combination.

[0181] A2, for each potential customer, determine the psychological price center and psychological price dispersion based on the multi-source price signals corresponding to the potential customer.

[0182] Among them, the psychological price center represents the price that the corresponding potential customer is most likely to accept, and the psychological price dispersion represents the degree of dispersion of the price that the corresponding potential customer can accept.

[0183] Specifically, the process of determining the psychological price center and psychological price dispersion can be performed separately for each potential customer in the potential customer set. For the first... For each potential customer, the multi-source price signals corresponding to that customer can be converted into a unified price unit, and the psychological price center can be determined based on the concentrated position of the effective price signals on the price axis. .

[0184] In one implementation, the potential customer's willingness to pay price determined by the first quantile aggregation in step S202 can be used as the psychological price center for the corresponding potential customer. Since the potential customer's willingness to pay price is obtained by aggregating the multi-source price signals corresponding to the corresponding potential customer, this psychological price center still belongs to the price parameter determined based on the multi-source price signals.

[0185] In other implementations, after filtering outliers from multi-source price signals, the psychological price center can be determined based on the central position of the retained price signals. The closer the psychological price center is to a certain price, the higher the relative likelihood that the corresponding potential customer will accept that price.

[0186] The dispersion of psychological prices can be expressed as ,and It is a positive number. The psychological price dispersion can be determined by the degree of dispersion of the multi-source price signals corresponding to the potential customer on both sides of the psychological price center. The more concentrated the multi-source price signals are, the smaller the psychological price dispersion is; the wider the price range covered by the multi-source price signals, the greater the psychological price dispersion is.

[0187] A smaller dispersion of psychological price indicates that potential customers have a narrower budget range and are more sensitive to price changes; a larger dispersion of psychological price indicates that potential customers have a wider acceptable price range and are relatively less sensitive to price changes.

[0188] When the number of valid price signals corresponding to a potential customer is insufficient to directly determine the psychological price dispersion, the psychological price dispersion of the corresponding potential customer can also be determined based on the price dispersion of the target SKU, the vehicle series to which the target SKU belongs, or the price dispersion of historical potential customers with the same potential customer characteristics.

[0189] A3 employs a psychological price density construction method, establishing individual psychological price densities for each potential customer based on their corresponding psychological price center and psychological price dispersion.

[0190] When the psychological price density is constructed using a Gaussian distribution, the first... The individual psychological price density of a potential customer can be expressed as: in, Indicates the first A potential customer at a price Individual psychological price density at a given location; Indicates the price to be evaluated; Indicates the first The psychological price center of potential customers; Indicates the first The dispersion of the psychological price of potential customers.

[0191] Individual psychological price density at the psychological price center The surrounding area has a high density value, and it increases with price. With the Psychological Price Center The psychological price density decreases as the distance between them increases. Therefore, individual psychological price density can reflect the relative acceptance level of potential customers at different price levels.

[0192] When the dispersion of psychological prices is small, the density of individual psychological prices is relatively concentrated, indicating that the price range acceptable to the corresponding potential customers is narrow; when the dispersion of psychological prices is large, the density of individual psychological prices is relatively dispersed, indicating that the price range acceptable to the corresponding potential customers is wide.

[0193] It should be noted that, This is a probability density function and does not imply that potential customers will prioritize price. The probability of purchasing a single target SKU.

[0194] A4. Based on the potential customer characteristic information corresponding to each potential customer, determine the corresponding potential customer weight, and based on the potential customer weight corresponding to each potential customer, weighted aggregate the individual psychological price density of each potential customer to obtain the group consensus price density of the target SKU.

[0195] Regarding the first For each potential customer, the potential customer weight can be determined based on at least one of the following: the corresponding behavior type, the timeliness of the behavior, and the user profile. .

[0196] For example, the weight of a potential customer who makes a purchase can be higher than that of one who leaves a contact's information, and the weight of one who leaves a contact's information can be higher than that of one who adds to favorites or clicks. For the same type of behavior, the closer the behavior occurs to the current calculation time, the higher the corresponding potential customer weight can be. The better the user profile matches the price positioning of the target SKU, the higher the corresponding potential customer weight can be.

[0197] Each potential customer's weight can be non-negative, and at least one potential customer's weight must be greater than zero. By weighted aggregation of the individual psychological price densities of each potential customer, the group consensus price density can be obtained: in, Indicates the target SKU in terms of price The consensus price density at a given location; This indicates the number of potential customers in the potential customer set; Indicates the first The potential customer weight of each potential customer.

[0198] By weighting individual psychological price density with potential customer weights, we can increase the influence of potential customers with strong purchase intentions, recent behavior, or user profiles that match the target SKU on the group consensus price density, while reducing the influence of potential customers with weak purchase intentions or earlier behavior on the group consensus price density.

[0199] It should be noted that, This represents the consensus price density, not the cumulative distribution function. When determining the adaptive recommendation price later, the consensus price density can be accumulated along the price-increasing direction to obtain the corresponding relationship between price and cumulative probability.

[0200] A5 defines the continuous price range in the group consensus price density where the density value is not lower than the density threshold as the theoretical pricing range for the target SKU.

[0201] Specifically, the consensus price density can be traversed along the direction of price increase, and continuous price ranges with density values ​​reaching or exceeding the density threshold can be identified as high-density price ranges. The lower and upper boundary prices of these high-density price ranges can be used as the two endpoints of the theoretical pricing interval.

[0202] When the consensus price density exhibits a unimodal distribution, the price range with a density value not lower than the density threshold can form a continuous price interval encompassing the peak of the consensus price density. This continuous price interval can characterize the price range where the psychological price distribution of multiple potential customers is relatively concentrated, and is thus determined as the theoretical pricing interval for the target SKU.

[0203] When the price range where the consensus price density exhibits a multi-peak distribution and the density value is not lower than the density threshold forms multiple consecutive price intervals that are separate from each other, the consecutive price interval containing the price corresponding to the maximum density value of the consensus price density can be determined as the theoretical pricing interval; alternatively, the cumulative density corresponding to each consecutive price interval can be determined separately, and the consecutive price interval with the largest cumulative density can be determined as the theoretical pricing interval.

[0204] For example, when the density threshold is set to 80% of the maximum density value of the consensus price density, the continuous price range in the consensus price density that reaches more than 80% of the maximum density height can be determined as the theoretical pricing range. This process is based on the density height of the consensus price density and does not involve selecting the 80th percentile on the cumulative distribution function.

[0205] It should be noted that the theoretical pricing range is determined by the weighted aggregation of individual psychological price densities, reflecting the psychological consensus of potential customers on the price axis; the guiding price range is determined by the aggregation of two-stage quantiles at the user and SKU granularities, reflecting the quantile range of multiple potential customers' willingness to pay price samples. The theoretical pricing range and the guiding price range are price ranges generated by two different paths, and they are not required to have the same lower limit, upper limit, or width.

[0206] In some embodiments, after establishing the consensus price density of the target SKU, an adaptive recommended price for the target SKU can be determined from the consensus price density by combining the market status and competitive status of the target SKU. This process may include the following B1 to B5.

[0207] It should be noted that B1 to B5 output an adaptive recommended price, while A1 to A5 output a theoretical pricing range, and steps S201 to S205 output a guiding price range. The output formats and determination paths of the three are different.

[0208] B1, accumulate the consensus price density along the price increase direction to obtain the correspondence between price and cumulative probability.

[0209] Specifically, starting from the low end of the preset price range, the consensus price density can be accumulated sequentially along the price increase direction, so that each price corresponds to a corresponding cumulative probability.

[0210] The consensus price density can be expressed as Used to characterize the target SKU in terms of price The consensus density at a given point. After accumulating the consensus price density, the corresponding cumulative distribution function can be obtained. .in, Indicates that the price is not higher than The cumulative probability corresponding to the time.

[0211] The relationship between price and cumulative probability can be stored using a continuous function, a discrete mapping table, or an ordered data sequence. When using a discrete mapping table, multiple price evaluation points can be set within a preset price range, arranged in ascending order. The density contribution corresponding to each price evaluation point is accumulated sequentially, and the accumulated result is normalized so that the cumulative probability corresponding to the last price evaluation point is 1.

[0212] As the price increases, the cumulative probability remains constant or increases. Therefore, the correspondence between price and cumulative probability can be used to reverse-engineer the corresponding price based on a specified quantile.

[0213] It should be noted that, For group consensus price density, The cumulative distribution function, obtained by accumulating the consensus price density, and the cumulative distribution function are different data objects. Subsequent quantile prices should be determined based on the cumulative distribution function, rather than by directly performing an inverse function operation on the consensus price density.

[0214] B2 obtains a pre-defined set of candidate percentiles and a competitive scoring method, as well as the market scarcity, same-store competition level, and exposure conversion efficiency of the target SKU.

[0215] The candidate quantile set can include multiple pre-defined candidate quantiles. Each candidate quantile is a value between 0 and 1, representing a cumulative probability position on the cumulative distribution function, rather than a specific price. The candidate quantile set can include the median quantile and other quantiles located on either side of the median quantile.

[0216] The competitive scoring method determines the degree of match between each candidate percentile and the current market status of the target SKU, based on factors such as market scarcity, same-store competition, and exposure-to-conversion efficiency. This method can pre-save scoring parameters or rules for each candidate percentile, allowing different candidate percentiles to receive different competitive scores under the same market conditions.

[0217] Market scarcity is used to characterize the degree of supply and demand tension of a target SKU in the current market. Market scarcity can be determined based on at least one of the following: market demand for the target SKU, available inventory, historical sales velocity, or supply-demand gap. A higher market scarcity indicates a greater market demand for the target SKU relative to its available supply.

[0218] Same-store competition level is used to characterize the degree of competition posed by other vehicles with the same or similar market positioning to the target SKU within the same dealership or sales entity. Same-store competition level can be determined based on at least one of the following: the number of similar models in the same store, the number of vehicles with similar prices in the same store, the degree of inventory overlap, or the exposure of similar vehicles.

[0219] Exposure-to-conversion efficiency (ECE) characterizes how efficiently exposure to a target SKU converts into subsequent actions such as clicks, favorites, lead generation, or sales. A higher EEC indicates a stronger ability of the target SKU to generate further interaction or purchase behavior from potential customers after gaining exposure.

[0220] In some embodiments, market scarcity, same-store competition, and exposure conversion efficiency can be normalized to enable market indicators of different dimensions to be used in competitive scoring methods.

[0221] B3 uses a competitive scoring method to determine the competitive score for each candidate percentile in the candidate percentile set based on market scarcity, same-store competition, and exposure conversion efficiency.

[0222] Among them, market scarcity and exposure conversion efficiency have a positive impact on the competition score, while the degree of same-store competition has a negative impact on the competition score.

[0223] Specifically, each candidate percentile, along with the market scarcity, same-store competition level, and exposure conversion efficiency of the target SKU, can be input into the competition scoring method to obtain the competition score corresponding to each candidate percentile.

[0224] The competitive scoring method can configure corresponding scoring parameters or market status matching rules for different candidate percentiles. For example, higher candidate percentiles can establish a higher degree of matching with market statuses that have higher market scarcity and higher exposure-to-conversion efficiency; lower candidate percentiles can establish a higher degree of matching with market statuses that have higher levels of same-store competition.

[0225] Market scarcity has a positive impact on competition scores because, all other things being equal, an increase in market scarcity leads to a higher competition score for candidate percentiles that match a higher recommend price preference. Similarly, exposure-to-conversion efficiency also has a positive impact on competition scores because, an increase in exposure-to-conversion efficiency leads to a higher competition score for candidate percentiles that match a higher recommend price preference.

[0226] The degree of same-store competition has a negative impact on competition scores. This means that, all other things being equal, when the degree of same-store competition increases, the competition scores corresponding to candidate percentiles that match the higher recommended price tendency decrease, thereby increasing the likelihood that lower candidate percentiles will be selected.

[0227] Therefore, when the target SKU is relatively scarce or the exposure conversion efficiency is high, the target percentile determined later can be moved to a higher percentile; when the level of competition among stores is high, the target percentile determined later can be moved to a lower percentile.

[0228] B4 performs Softmax processing on the competitive scores corresponding to each candidate quantile and determines the target quantile from the candidate quantile set based on the processing results.

[0229] Softmax processing is used to convert the competitive scores corresponding to each candidate quantile into normalized results. The processing results for each candidate quantile are non-negative, and the sum of the processing results for all candidate quantiles is 1. The higher the competitive score, the larger the processing result corresponding to the candidate quantile after Softmax processing.

[0230] Specifically, the competitive score can be determined one by one according to the candidate percentiles. For each candidate percentile, the market scarcity, same-store competition level, and exposure conversion efficiency are combined according to the scoring parameters or market status matching rules pre-configured for that candidate percentile to obtain the competitive score corresponding to that candidate percentile.

[0231] Different candidate percentiles can be configured with different scoring parameters or market status matching rules, so that candidate percentiles under the same market status receive different competitive scores. Market scarcity and exposure-to-conversion efficiency are used to improve the competitive score of candidate percentiles that match the higher recommended price tendency, while same-store competition is used to reduce the competitive score of such candidate percentiles and relatively increase the likelihood of lower candidate percentiles being selected.

[0232] When multiple candidate quantiles are set, corresponding competitive scores can be obtained based on the scoring parameters of each candidate quantile, and Softmax processing can be performed on multiple competitive scores. The candidate quantile with the largest Softmax processing result can be determined as the target quantile.

[0233] In other implementations, the target quantile can be determined from the candidate quantile set by probabilistic selection based on the Softmax processing results corresponding to each candidate quantile.

[0234] B5. Based on the correspondence between price and cumulative probability, determine the price corresponding to the cumulative probability reaching the target percentile, and set this price as the adaptive recommendation price for the target SKU.

[0235] Specifically, based on the correspondence between the price and the cumulative probability obtained in step B1, the target quantile for the cumulative probability can be found. The corresponding price at that time. The adaptive recommended price for the target SKU can be represented as follows: in, This represents the adaptive recommended price for the target SKU; This represents the inverse cumulative distribution determination operation corresponding to the relationship between price and cumulative probability; Indicates the target quantile.

[0236] When the correspondence between price and cumulative probability is stored using a discrete mapping table or an ordered data sequence, the search can be performed along the direction of increasing price, and the price at which the cumulative probability first reaches or exceeds the target quantile is determined as the adaptive recommended price.

[0237] When the relationship between price and cumulative probability is represented by a continuous function, the corresponding quantile price can be determined based on the target quantile on the cumulative distribution function, and this quantile price can be used as the adaptive recommendation price.

[0238] It should be noted that the inverse function in the formula is the cumulative distribution function. Rather than the price density of group consensus The consensus price density needs to be processed by B1 to obtain the correspondence between price and cumulative probability before the corresponding price can be determined based on the target quantile.

[0239] By dynamically selecting the target percentile based on market scarcity, same-store competition, and exposure-to-conversion efficiency, the practice of always using a fixed percentile to determine the recommended price can be avoided. When the market scarcity, exposure-to-conversion efficiency, or same-store competition of the target SKU changes, the target percentile and adaptive recommended price can be adjusted accordingly, thus ensuring that the adaptive recommended price is adapted to the current market and competitive situation of the target SKU.

[0240] In some embodiments, steps B1 to B5 can be executed at multiple calculation times throughout the day. For each calculation time, the market scarcity, same-store competition level, and exposure conversion efficiency corresponding to that calculation time can be obtained, and the target percentile and adaptive recommendation price corresponding to that calculation time can be determined based on the relevant market indicators.

[0241] Because the inventory status, market demand, same-store competition, and exposure-to-conversion rates of a target SKU can change over time, the target percentile determined at different calculation points can differ, and the corresponding adaptive recommendation price can also change over time. The adaptive recommendation prices determined at multiple calculation points can be arranged in chronological order to form an adaptive recommendation price sequence.

[0242] It can retrieve multiple adaptive recommended prices within a preset time range, and determine the lower and upper limits of the adaptive recommended price range based on the lower and higher prices, respectively, thereby generating the adaptive recommended price range for the target SKU within that preset time range. The preset time range can be one day, multiple hours, or a sliding time window set according to business needs.

[0243] In some implementations, the lowest and highest adaptive recommended prices within a preset time range can be used as the two endpoints of the adaptive recommended price interval. In other implementations, outlier filtering can be applied to multiple adaptive recommended prices within a preset time range, and the two endpoints of the adaptive recommended price interval can be determined based on the lower and higher prices retained after filtering, thus reducing the impact of abnormal market indicators at individual moments on the interval endpoints.

[0244] When a new calculation time arrives, the newly determined adaptive recommendation price can be added to the adaptive recommendation price sequence, and the adaptive recommendation price range can be updated based on the updated adaptive recommendation price sequence. Therefore, the adaptive recommendation price range can dynamically change with the market and competitive status of the target SKU.

[0245] It should be noted that at a single calculation moment, the adaptive recommended price output by B5 can be a single price point; after aggregating the adaptive recommended prices from multiple calculation moments within a preset time range, an adaptive recommended price range can also be output. This adaptive recommended price range reflects the range of changes in the recommended price over time, and is not equivalent to the theoretical pricing range determined by A1 to A5, nor is it equivalent to the guiding price range determined by steps S201 to S205.

[0246] The vehicle price range prediction method provided in this application first obtains the target SKU, the set of potential customers corresponding to the target SKU, and the multi-source price signals corresponding to each potential customer. Then, for each potential customer, the first quantile aggregation is performed on the corresponding multi-source price signals at the user granularity to obtain the potential customer's willingness to pay price, and a sample set of SKU willingness to pay prices for the target SKU is constructed based on the potential customer's willingness to pay prices. Next, the second quantile aggregation is performed on the SKU willingness to pay price sample set at the SKU granularity to obtain the lower limit and upper limit of the guidance price range for the target SKU. Finally, the guidance price range for the target SKU is generated based on the lower limit and upper limit of the guidance price range. This application, by obtaining price signals from different sources, can increase the data sources for vehicle price analysis and reduce the impact of insufficient data volume on the prediction results when using only a single source of price data. By first aggregating multi-source price signals corresponding to the same potential customer at the user granularity, each potential customer participates in subsequent processing with their corresponding willingness-to-pay price. This reduces data bias caused by duplicate inclusion of multiple price signals from the same potential customer, and makes the resulting willingness-to-pay price more fully represent the price acceptance of different potential customers for the target SKU. Further, by aggregating the willingness-to-pay prices of each potential customer at the SKU granularity, the impact of individual abnormal prices on the endpoints of the price range can be reduced, resulting in a more stable and reasonable range width for the generated guidance price range, thereby improving its actual reference value.

[0247] Further reference Figure 5 As an implementation of the methods shown in the above figures, this application provides an embodiment of a vehicle price range prediction device, which is similar to... Figure 2 Corresponding to the method embodiments shown, this device can be specifically applied to various terminal devices and servers.

[0248] like Figure 5 As shown, the vehicle price range prediction device of this embodiment, device 500 includes: The system includes a data acquisition unit 501, a first quantile aggregation unit 502, a sample set construction unit 503, a second quantile aggregation unit 504, and an interval generation unit 505.

[0249] Data acquisition unit 501 is used to acquire the target SKU, the set of potential customers corresponding to the target SKU, and the multi-source price signals corresponding to each potential customer in the potential customer set, wherein the multi-source price signals include price signals from at least two different sources. First quantile aggregation unit 502 is used to perform first quantile aggregation on the corresponding multi-source price signals at the user granularity for each potential customer, obtaining the potential customer's willingness to pay price. Sample set construction unit 503 is used to construct a sample set of SKU payment willingness prices for the target SKU based on the potential customer's willingness to pay price, wherein the SKU payment willingness price sample set includes at least one potential customer's willingness to pay price. Second quantile aggregation unit 504 is used to perform second quantile aggregation on the SKU payment willingness price sample set at the SKU granularity, obtaining the lower limit and upper limit of the guidance price range for the target SKU. Range generation unit 505 is used to generate the guidance price range for the target SKU based on the lower limit and upper limit of the guidance price range.

[0250] In this embodiment, the specific processing of the data acquisition unit 501, the first quantile aggregation unit 502, the sample set construction unit 503, the second quantile aggregation unit 504, and the interval generation unit 505, and the resulting technical effects, can be found in reference to [the relevant documentation]. Figure 2 The relevant descriptions of steps S201 to S205 in the corresponding embodiments will not be repeated here.

[0251] In some alternative implementations, the first quantile aggregation unit 502 may specifically be used for: Arrange the multi-source price signals corresponding to the potential customers in ascending order of price values ​​to obtain an ordered price signal sequence.

[0252] Based on the preset lower and upper truncation quantiles, the lower and upper quantile thresholds of the ordered price signal sequence are determined respectively.

[0253] Retain price signals in an ordered price signal sequence whose price values ​​are neither lower than the lower quantile threshold nor higher than the upper quantile threshold.

[0254] The median of the retained price signals is determined as the potential customer's willingness to pay price.

[0255] In some alternative implementations, the first quantile aggregation unit 502 may specifically be used for: Obtain the signal source type and generation time of each price signal corresponding to the potential customer, as well as the preset source credibility determination rules, exponential time decay method, and target weighted quantile.

[0256] The source reliability of each price signal is determined based on the signal source type and source reliability determination rules.

[0257] The time interval corresponding to each price signal is determined based on the generation time of each price signal and the current calculation time.

[0258] The exponential time decay method is used to perform time decay processing on the source reliability of each price signal according to the time interval corresponding to each price signal, so as to obtain the signal record weight of each price signal.

[0259] Arrange the multi-source price signals corresponding to the potential customers in ascending order of price values ​​to obtain an ordered price signal sequence.

[0260] The total weight of the signal records is obtained by summing the signal record weights of each price signal in the ordered price signal sequence.

[0261] According to the order of the ordered price signal sequence, for each price signal, the signal record weights corresponding to the price signal and the price signals preceding it are accumulated to obtain the cumulative weight of the price signal. Based on the ratio of the cumulative weight to the total weight of the signal records, the normalized cumulative weight of the price signal is determined.

[0262] From the ordered price signal sequence, select the price signal whose normalized cumulative weight first reaches or exceeds the target weighted quantile, and determine the price value corresponding to the selected price signal as the potential customer's willingness to pay price.

[0263] In some alternative implementations, the second quantile aggregation unit 504 may specifically be used for: Obtain the preset lower and upper quantiles.

[0264] Arrange the potential customers' willingness to pay prices in the SKU willingness-to-pay price sample set in ascending order of price value to obtain an ordered willingness-to-pay price sequence.

[0265] The lower limit of the guidance price range is determined from the ordered willingness-to-pay price sequence based on the lower quantile.

[0266] Based on the upper quantile, the upper limit of the guidance price range is determined from the ordered willingness-to-pay price sequence.

[0267] In some optional embodiments, the device further includes a psychological price parameter determination unit (not shown), an individual psychological price density construction unit (not shown), a group consensus price density determination unit (not shown), and a theoretical pricing range determination unit (not shown).

[0268] The data acquisition unit is also used to acquire a preset psychological price density construction method and density threshold, as well as potential customer characteristic information corresponding to each potential customer in the potential customer set. The potential customer characteristic information includes at least one of behavior type, behavior timeliness, and user profile.

[0269] The psychological price parameter determination unit is used to determine the psychological price center and psychological price dispersion for each potential customer based on the multi-source price signals corresponding to the potential customer. The psychological price center represents the price that the corresponding potential customer is most likely to accept, and the psychological price dispersion represents the degree of dispersion of the price that the corresponding potential customer can accept.

[0270] The individual psychological price density construction unit is used to establish the individual psychological price density of each potential customer based on the psychological price center and psychological price dispersion of each potential customer using the psychological price density construction method.

[0271] The group consensus price density determination unit is used to determine the corresponding potential customer weight based on the potential customer characteristic information of each potential customer, and to perform weighted aggregation of the individual psychological price density of each potential customer based on the potential customer weight, so as to obtain the group consensus price density of the target SKU.

[0272] The theoretical pricing range determination unit is used to determine the theoretical pricing range of the target SKU as a continuous price range in the group consensus price density where the density value is not lower than the density threshold.

[0273] In some alternative implementations, the apparatus further includes an accumulation processing unit (not shown), a competitive scoring determination unit (not shown), a target percentile determination unit (not shown), and an adaptive recommendation price determination unit (not shown).

[0274] The cumulative processing unit is used to accumulate the consensus price density along the price increase direction to obtain the correspondence between price and cumulative probability.

[0275] The data acquisition unit is also used to acquire a preset set of candidate quantiles and a competitive scoring method, as well as the market scarcity, same-store competition level, and exposure conversion efficiency of the target SKU.

[0276] The competition score determination unit is used to determine the competition score corresponding to each candidate percentile in the candidate percentile set based on the competition score determination method, according to market scarcity, same-store competition level, and exposure conversion efficiency. Among them, market scarcity and exposure conversion efficiency have a positive impact on the competition score, while same-store competition level has a negative impact on the competition score.

[0277] The target quantile determination unit is used to perform Softmax processing on the competitive scores corresponding to each candidate quantile and determine the target quantile from the candidate quantile set based on the processing results.

[0278] The adaptive recommendation price determination unit is used to determine the price corresponding to the cumulative probability reaching the target percentile based on the correspondence between price and cumulative probability, and to determine the price as the adaptive recommendation price of the target SKU.

[0279] It should be noted that the implementation details and technical effects of each unit in the vehicle price range prediction device provided in the embodiments of this application can be referred to the descriptions of other embodiments in this application, and will not be repeated here.

[0280] The following is for reference. Figure 6 It shows a schematic diagram of the structure of a computer system 600 suitable for implementing electronic devices according to embodiments of the present application. Figure 6 The computer system 600 shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0281] like Figure 6 As shown, the computer system 600 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage device 608 into a random access memory (RAM) 603. The RAM 603 also stores various programs and data required for the operation of the computer system 600. The processing device 601, ROM 602, and RAM 603 are interconnected via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0282] Typically, the following devices can be connected to I / O interface 605: input devices 606 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, etc.; output devices 607 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 608 including, for example, magnetic tapes, hard disks, etc.; and communication devices 609. Communication device 609 allows computer system 600 to communicate wirelessly or wiredly with other devices to exchange data such as target SKUs, potential customer data, multi-source price signals, and guidance price ranges. Although Figure 6 A computer system 600 with various electronic devices is shown, but it should be understood that implementation or possession of such devices is not required. Figure 6 All the devices shown may be implemented alternatively or have more or fewer devices.

[0283] Specifically, according to embodiments of this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the vehicle price range prediction method shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 609, or installed from a storage device 608, or from a ROM 602. When the computer program is executed by a processing device 601, it performs the corresponding functions defined in the vehicle price range prediction method of the embodiments of this application.

[0284] It should be noted that the computer-readable medium in this application can be a computer-readable signal medium, a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by, or in conjunction with, an instruction execution system, apparatus, or device.

[0285] In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, wherein computer-readable program code is carried on the data signal. The propagated data signal may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the foregoing. The computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which may transmit, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium may be transmitted via any suitable medium, including but not limited to wires, optical cables, radio frequency (RF) signals, or any suitable combination of the foregoing media.

[0286] The aforementioned computer-readable medium may be included in the aforementioned electronic device, or it may exist independently without being assembled in the electronic device.

[0287] The aforementioned computer-readable medium carries one or more programs. When the aforementioned one or more programs are executed by an electronic device, the electronic device performs the following functions: Figure 2 The illustrated embodiments and their alternative implementations demonstrate a method for predicting vehicle price ranges.

[0288] Computer program code for performing the operations of this application may be written in one or more programming languages ​​or a combination thereof. Programming languages ​​may include object-oriented programming languages ​​such as Java, Smalltalk, C++, and Python, as well as conventional procedural programming languages ​​such as C or similar languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN); the remote computer may also be connected to an external computer, for example, via the Internet using an Internet service provider.

[0289] The flowcharts and block diagrams in the accompanying drawings illustrate possible implementation architectures, functions, and operations of systems, methods, and computer program products according to different embodiments of this application. Each block in a flowchart or block diagram may represent a module, program segment, or portion of code containing one or more executable instructions for implementing the corresponding logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may be executed in a different order than shown in the drawings. For example, two consecutively shown blocks may actually be executed substantially in parallel, or they may be executed in reverse order depending on the functions involved. Each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented by a dedicated hardware system that performs the corresponding function or operation, or by a combination of dedicated hardware and computer instructions.

[0290] The modules involved in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the module itself in certain circumstances. For example, the data acquisition unit can also be described as "a unit for acquiring a target SKU, a set of potential customers corresponding to the target SKU, and multi-source price signals corresponding to each potential customer in the set of potential customers."

[0291] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of protection of this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the concept of this application. For example, technical solutions formed by substituting the above-described technical features with similar functions disclosed in this application.

Claims

1. A method for predicting vehicle price ranges, characterized in that, The method includes: Obtain the target SKU, the potential customer set corresponding to the target SKU, and the multi-source price signal corresponding to each potential customer in the potential customer set, wherein the multi-source price signal includes price signals from at least two different sources; For each potential customer, the first quantile of the corresponding multi-source price signals is aggregated at the user granularity to obtain the potential customer's willingness to pay price. Based on the potential customer's willingness to pay price, construct a sample set of SKU willingness to pay prices for the target SKU, wherein the sample set of SKU willingness to pay prices includes at least one potential customer willingness to pay price; At the SKU granularity, the second quantile aggregation is performed on the sample set of the SKU's willingness to pay price to obtain the lower limit and upper limit of the guidance price range for the target SKU; The guidance price range for the target SKU is generated based on the lower limit and the upper limit of the guidance price range.

2. The method according to claim 1, characterized in that, The step of aggregating the corresponding multi-source price signals at the user granularity using the first quantile to obtain the potential customer's willingness to pay price includes: Arrange the multi-source price signals corresponding to the potential customers in ascending order of price values ​​to obtain an ordered price signal sequence; Based on the preset lower truncation quantile and upper truncation quantile, the lower quantile threshold and upper quantile threshold of the ordered price signal sequence are determined respectively; Retain price signals in the ordered price signal sequence whose price values ​​are not lower than the lower quantile threshold and not higher than the upper quantile threshold; The median of the retained price signals is determined as the potential customer's willingness to pay price.

3. The method according to claim 1, characterized in that, The step of aggregating the corresponding multi-source price signals at the user granularity using the first quantile to obtain the potential customer's willingness to pay price includes: Obtain the signal source type and generation time of each price signal corresponding to the relevant potential customer, as well as the preset source credibility determination rules, exponential time decay method and target weighted quantile; The source reliability of each price signal is determined based on the signal source type and the source reliability determination rules. The time interval corresponding to each price signal is determined based on the generation time of each price signal and the current calculation time. Using the exponential time decay method, the source reliability of each price signal is time decayed according to the time interval corresponding to each price signal to obtain the signal recording weight of each price signal. Arrange the multi-source price signals corresponding to the potential customers in ascending order of price values ​​to obtain an ordered price signal sequence; The total weight of the signal records is obtained by summing the signal record weights of each price signal in the ordered price signal sequence. According to the order of the ordered price signal sequence, for each price signal, the signal record weights corresponding to the price signal and the price signals arranged before it are accumulated to obtain the cumulative weight corresponding to the price signal. Based on the ratio of the cumulative weight to the total weight of the signal records, the normalized cumulative weight corresponding to the price signal is determined. From the ordered price signal sequence, select the price signal whose normalized cumulative weight first reaches or exceeds the target weighted quantile, and determine the price value corresponding to the selected price signal as the potential customer's willingness to pay price.

4. The method according to claim 1, characterized in that, The step of performing second quantile aggregation on the SKU-level sample set of willingness-to-pay prices to obtain the lower limit and upper limit of the guidance price range for the target SKU includes: Obtain the preset lower and upper quantiles; Arrange the potential customers' willingness to pay prices in the SKU willingness-to-pay price sample set in ascending order of price value to obtain an ordered willingness-to-pay price sequence; The lower limit of the guiding price range is determined from the ordered willingness-to-pay price sequence based on the lower quantile. The upper limit of the guidance price range is determined from the ordered willingness-to-pay price sequence based on the upper quantile.

5. The method according to claim 1, characterized in that, The method further includes: Obtain a preset psychological price density construction method and density threshold, as well as potential customer feature information corresponding to each potential customer in the potential customer set, wherein the potential customer feature information includes at least one of behavior type, behavior timeliness and user profile; For each potential customer, the psychological price center and psychological price dispersion are determined based on the multi-source price signals corresponding to the potential customer. The psychological price center represents the price that the potential customer is most likely to accept, and the psychological price dispersion represents the degree of dispersion of the price that the potential customer can accept. Using the aforementioned psychological price density construction method, individual psychological price densities are established for each potential customer based on their corresponding psychological price center and psychological price dispersion. Based on the potential customer characteristic information corresponding to each potential customer, the corresponding potential customer weight is determined, and based on the potential customer weight corresponding to each potential customer, the individual psychological price density of each potential customer is weighted and aggregated to obtain the group consensus price density of the target SKU. The continuous price range in the consensus price density where the density value is not lower than the density threshold is determined as the theoretical pricing range of the target SKU.

6. The method according to claim 5, characterized in that, The method further includes: The consensus price density of the group is accumulated along the direction of price increase to obtain the correspondence between price and cumulative probability. Obtain a preset set of candidate quantiles and a competition scoring method, as well as the market scarcity, same-store competition level, and exposure conversion efficiency of the target SKU; Using the aforementioned competition score determination method, the competition score corresponding to each candidate percentile in the candidate percentile set is determined based on the market scarcity, the same-store competition level, and the exposure conversion efficiency. The market scarcity and the exposure conversion efficiency have a positive impact on the competition score, while the same-store competition level has a negative impact on the competition score. Softmax processing is performed on the competitive scores corresponding to each candidate quantile, and the target quantile is determined from the candidate quantile set based on the processing results; Based on the correspondence between price and cumulative probability, the price corresponding to the cumulative probability reaching the target percentile is determined, and the price is determined as the adaptive recommended price of the target SKU.

7. A vehicle price range prediction device, characterized in that, The device includes: The data acquisition unit is used to acquire a target SKU, a set of potential customers corresponding to the target SKU, and a multi-source price signal corresponding to each potential customer in the set of potential customers, wherein the multi-source price signal includes price signals from at least two different sources. The first quantile aggregation unit is used to perform first quantile aggregation on the corresponding multi-source price signals at the user granularity for each potential customer, so as to obtain the potential customer's willingness to pay price. The sample set construction unit is used to construct a sample set of SKU payment willingness prices for the target SKU based on the potential customer's corresponding potential customer payment willingness price, wherein the sample set of SKU payment willingness prices includes at least one potential customer payment willingness price; The second quantile aggregation unit is used to perform second quantile aggregation on the SKU payment willingness price sample set at the SKU granularity to obtain the lower limit and upper limit of the guidance price range of the target SKU; The interval generation unit is used to generate the guidance price range for the target SKU based on the lower limit and the upper limit of the guidance price range.

8. An electronic device, characterized in that, include: One or more processors; Storage device, on which one or more programs are stored, When one or more programs are executed by one or more processors, the one or more processors implement the method as claimed in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, It stores a computer program thereon, wherein the computer program, when executed by one or more processors, implements the method as claimed in any one of claims 1-6.

10. A computer program product, characterized in that, Includes a computer program / instruction that, when executed by a processor, implements the method as described in any one of claims 1-6.

Images