Method, device, equipment and storage medium for identifying abnormal product bids

Abstract

The present application provides a method, device, equipment and storage medium for identifying abnormal bids for commodities, wherein the method includes: obtaining real-time features of target commodities, and creating a picture feature set and a text vector set, extracting picture feature information and title vectors of target commodities from bid logs, and determining picture similarity feature sequences, picture trend feature sequences, text similarity feature sequences and title trend feature sequences of target commodities, and inputting picture similarity feature sequences, picture trend feature sequences, text similarity feature sequences, title trend feature sequences and real-time features of target commodities into a bid recognition model obtained by pre-training to obtain bid recognition results of target commodities. The method of the present application improves the accuracy and reliability of bid recognition by using multimodal information for bid recognition and judgment.

Description

Technical Field

[0001] The present application relates to the field of e-commerce technology, and in particular to a method, device, equipment and storage medium for identifying abnormal bidding for commodities. Background Art

[0002] In e-commerce platforms, merchants and users can bid for goods, but since the prices of goods are subject to external influences and will fluctuate greatly, when the price of goods fluctuates abnormally and does not conform to the mainstream price range, there will be a phenomenon of chaotic commodity prices, which disrupts the normal trading order. Therefore, it is necessary to identify the bids of goods to ensure that the bids of goods are within a reasonable range.

[0003] Currently, bid identification for commodities mainly uses time series or machine learning methods to predict price ranges, and then performs bid identification based on the predicted price ranges.

[0004] However, the prior art methods have the problem of inaccurate prediction results, which in turn leads to inaccurate bid identification results. Summary of the invention

[0005] The purpose of this application is to provide a method, device, equipment and storage medium for identifying abnormal bids for commodities in view of the deficiencies in the above-mentioned prior art, so as to solve the problem of inaccurate bid identification results in the prior art.

[0006] To achieve the above objectives, the technical solutions adopted in this application are as follows:

[0007] In a first aspect, the present application provides a method for identifying abnormal bids for commodities, the method comprising:

[0008] Acquire real-time features of a target commodity, and create a picture feature set and a text vector set according to community dynamic information, wherein the real-time features are used to characterize transaction information of the target commodity, and the community dynamic information is used to record pictures and text information of multiple commodities, wherein the picture feature set includes multiple picture features, and the text vector set includes multiple text vectors;

[0009] Extracting the image feature information of the target product from the bidding log, and determining the image similarity feature sequence and the image trend feature sequence of the target product according to the image feature information and the image feature set, wherein the bidding log is used to record the bidding information of the target product, the image similarity feature sequence includes a plurality of sequentially arranged quantities, each of which represents the number of image features similar to the image feature information in each time period within a preset time interval, and the image trend feature sequence includes a plurality of sequentially arranged trend features, each of which is used to characterize the change information of the number of similar image features in each time period within the preset time interval relative to the previous time period;

[0010] Extracting the title vector of the target product from the bid log, and determining the text similarity feature sequence and the title trend feature sequence of the target product according to the title vector and the text vector set;

[0011] The image similarity feature sequence, image trend feature sequence, text similarity feature sequence, title trend feature sequence and real-time features of the target product are input into a pre-trained bid recognition model to obtain a bid recognition result of the target product in the bid log.

[0012] Optionally, determining the image similarity feature sequence and the image trend feature sequence of the target product according to the image feature information and the image feature set includes:

[0013] Determine the similarity between the picture feature information and each picture feature in the picture feature set, and obtain the picture similarity between each picture feature in the picture feature set and the picture feature information;

[0014] The picture similarity feature sequence and the picture trend feature sequence are determined based on a preset picture similarity threshold and the picture similarity between each picture feature and the picture feature information.

[0015] Optionally, the determining the picture similarity feature sequence and the picture trend feature sequence based on a preset picture similarity threshold and the picture similarity between each picture feature and the picture feature information includes:

[0016] Screening the picture features in the picture feature set according to the similarity threshold and the picture similarity between each picture feature and the picture feature information to obtain a plurality of target picture features;

[0017] Obtaining the picture similarity feature sequence according to the number of target picture features, the time corresponding to each target picture feature, and the preset time interval;

[0018] Perform trend calculation on the picture similarity feature sequence to obtain the picture trend feature sequence.

[0019] Optionally, determining the text similarity feature sequence and the title trend feature sequence of the target product according to the title vector and the text vector set includes:

[0020] Determine the similarity between the title vector and each text vector in the text vector set, and obtain the text similarity between each text vector in the text vector set and the title vector;

[0021] Based on a preset text similarity threshold and the text similarity between each text vector and the title vector, the text similarity feature sequence and the title trend feature sequence are determined.

[0022] Optionally, determining the similarity between the title vector and each text vector in the text vector set to obtain the text similarity between each text vector in the text vector set and the title vector includes:

[0023] Determine the similarity between the word vector of the first word in the title vector and the word vectors of each word in the first text vector, and determine the maximum similarity of each word vector in the first text vector, and use the maximum similarity as the similarity between the first word and the first text vector, the first word is any word in the title vector, and the first text vector is any text vector in the text vector set;

[0024] The average value of the similarities between each word in the title vector and the first text vector is used as the similarity between the title vector and the first text vector.

[0025] Optionally, after obtaining the bid recognition result of the target product, the method further includes:

[0026] The bid of the target product is intercepted or outputted according to the bid identification result of the target product.

[0027] Optionally, before inputting the picture similarity feature sequence, the picture trend feature sequence, the text similarity feature sequence, the title trend feature sequence and the real-time feature of the target product into a pre-trained price recognition model to obtain the bid recognition result of the target product, the process includes:

[0028] Acquire sample real-time features of the sample target product, and create a sample image feature set and a sample text vector set according to the sample bidding log, wherein the sample bidding log includes: a preset sample bidding tag;

[0029] Extracting sample image feature information of the sample target product from the sample bidding log, and determining a sample image similarity feature sequence and a sample image trend feature sequence of the sample target product according to the sample image feature information and the sample image feature set;

[0030] Extracting a sample title vector of the sample target product from the sample bidding log, and determining a sample text similarity feature sequence and a sample title trend feature sequence of the sample target product according to the sample title vector and the sample text vector set;

[0031] The price recognition model is obtained by training a price binary classification model using a multi-classification tree according to the sample image similarity feature sequence, the sample image trend feature sequence, the sample text similarity feature sequence, the sample title trend feature sequence and the sample real-time features of the sample target commodity.

[0032] In a second aspect, the present application provides a device for identifying abnormal bidding for a commodity, the device comprising:

[0033] An acquisition module, used to acquire real-time features of a target commodity, and to create a picture feature set and a text vector set according to community dynamic information, wherein the real-time features are used to characterize transaction information of the target commodity, and the community dynamic information is used to record pictures and text information of multiple commodities, wherein the picture feature set includes multiple picture features, and the text vector set includes multiple text vectors;

[0034] An image extraction module is used to extract image feature information of a target product from a bidding log, and determine an image similarity feature sequence and an image trend feature sequence of the target product according to the image feature information and the image feature set, wherein the bidding log is used to record the bidding information of the target product, the image similarity feature sequence includes a plurality of sequentially arranged quantities, each of which represents the number of image features similar to the image feature information in each time period within a preset time interval, and the image trend feature sequence includes a plurality of sequentially arranged trend features, each of which is used to characterize change information of the number of similar image features in each time period within the preset time interval relative to the previous time period;

[0035] A text extraction module, used to extract the title vector of the target product from the bid log, and determine the text similarity feature sequence and the title trend feature sequence of the target product according to the title vector and the text vector set;

[0036] The recognition module is used to input the image similarity feature sequence, image trend feature sequence, text similarity feature sequence, title trend feature sequence and the real-time features of the target product into a pre-trained bid recognition model to obtain the bid recognition result of the target product in the bid log.

[0037] Optionally, the image extraction module is specifically used for:

[0038] Determine the similarity between the picture feature information and each picture feature in the picture feature set, and obtain the picture similarity between each picture feature in the picture feature set and the picture feature information;

[0039] The picture similarity feature sequence and the picture trend feature sequence are determined based on a preset picture similarity threshold and the picture similarity between each picture feature and the picture feature information.

[0040] Optionally, the image extraction module is further specifically used for:

[0041] Screening the picture features in the picture feature set according to the similarity threshold and the picture similarity between each picture feature and the picture feature information to obtain a plurality of target picture features;

[0042] Obtaining the picture similarity feature sequence according to the number of target picture features, the time corresponding to each target picture feature, and the preset time interval;

[0043] Perform trend calculation on the picture similarity feature sequence to obtain the picture trend feature sequence.

[0044] Optionally, the text extraction module is specifically used for:

[0045] Determine the similarity between the title vector and each text vector in the text vector set, and obtain the text similarity between each text vector in the text vector set and the title vector;

[0046] Based on a preset text similarity threshold and the text similarity between each text vector and the title vector, the text similarity feature sequence and the title trend feature sequence are determined.

[0047] Optionally, the text extraction module is further specifically used for:

[0048] Determine the similarity between the word vector of the first word in the title vector and the word vectors of each word in the first text vector, and determine the maximum similarity of each word vector in the first text vector, and use the maximum similarity as the similarity between the first word and the first text vector, the first word is any word in the title vector, and the first text vector is any text vector in the text vector set;

[0049] The average value of the similarities between each word in the title vector and the first text vector is used as the similarity between the title vector and the first text vector.

[0050] In a third aspect, the present application provides an electronic device comprising: a processor, a storage medium and a bus, wherein the storage medium stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor communicates with the storage medium via the bus, and the processor executes the machine-readable instructions to perform the steps of the method described in the first aspect above.

[0051] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the method described in the first aspect are executed.

[0052] The beneficial effects of the present application are as follows: by combining image similarity feature sequences, image trend feature sequences, text similarity feature sequences, title trend feature sequences and real-time features of target products for product bid recognition, the popularity trend of the target product in the recent period can be determined more accurately, and the impact of the external factor of the product popularity trend on the product price is taken into account, and the time series features constructed by multimodal images and structured data are introduced into the judgment of bid anomalies, thereby improving the accuracy of identifying abnormal product bids and avoiding the problem of low accuracy caused by price prediction based solely on the time series price of the product. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0054] Figure 1 A schematic diagram showing an application scenario provided by an embodiment of the present application is shown;

[0055] Figure 2 A flowchart of a method for identifying abnormal bidding for a commodity provided in an embodiment of the present application is shown;

[0056] Figure 3 A flow chart of determining a picture similarity feature sequence and a picture trend feature sequence provided by an embodiment of the present application is shown;

[0057] Figure 4A flowchart of another method of determining a picture similarity feature sequence and a picture trend feature sequence provided by an embodiment of the present application is shown;

[0058] Figure 5 A flowchart of determining a text similarity feature sequence and a title trend feature sequence provided by an embodiment of the present application is shown;

[0059] Figure 6 A flowchart for determining text similarity provided by an embodiment of the present application is shown;

[0060] Figure 7 A flowchart of a training bid recognition model provided by an embodiment of the present application is shown;

[0061] Figure 8 A schematic diagram of the structure of a device for identifying abnormal bidding of a commodity provided in an embodiment of the present application is shown;

[0062] Fig. 9 A schematic structural diagram of an electronic device provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0063] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. It should be understood that the drawings in the present application only serve the purpose of explanation and description and are not used to limit the scope of protection of the present application. In addition, it should be understood that the schematic drawings are not drawn in real proportion. The flowchart used in this application shows the operations implemented according to some embodiments of the present application. It should be understood that the operations of the flowchart can be implemented out of sequence, and the steps without logical context can be reversed in order or implemented simultaneously. In addition, those skilled in the art can add one or more other operations to the flowchart under the guidance of the content of the present application, or remove one or more operations from the flowchart.

[0064] In addition, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present application.

[0065] It should be noted that the term "comprising" will be used in the embodiments of the present application to indicate the existence of the features declared thereafter, but does not exclude the addition of other features.

[0066] Currently, bid identification for commodities mainly uses time series or machine learning methods to predict prices, and then performs bid identification based on the predicted price range.

[0067] Taking the use of time series for price range prediction as an example, it is mainly based on the historical price changes and cyclical characteristics of commodities. However, in actual scenarios, the cyclical changes in prices are complex, and it is difficult to find cyclical patterns in most cases. In addition, commodity prices are affected by many factors, and the bids of various merchants are uneven. Therefore, it is very difficult to predict the price range of commodities, and the prediction accuracy is low, which will lead to inaccurate bid recognition results.

[0068] Based on the above problems, the present application proposes a method for identifying abnormal bids for goods, which combines the price information of the goods and the popularity trend information of the goods on the platform to perform multimodal information judgment, so as to more accurately identify abnormal bids for goods.

[0069] The method of this application can be used to supervise bidding behavior on e-commerce platforms, such as Figure 1 As shown, this is an application scenario of the method provided by the present application. When a user bids for product A, the electronic device can obtain the bidding log of product A, and identify the bid for product A based on the bidding log of product A and previous community dynamic information, and process the user's bidding behavior for product A based on the bid identification result. For example, when the bid identification result indicates that the bid is abnormal, the user's bidding behavior can be intercepted.

[0070] Next, combine Figure 2 , the method for identifying abnormal bidding of commodities in the present application is further described, and the execution subject of the method can be an electronic device, such as Figure 2 As shown, the method includes:

[0071] S201: Acquire the real-time features of the target product, and create a picture feature set and a text vector set based on the community dynamic information. The real-time features are used to characterize the transaction information of the target product, and the community dynamic information is used to record the pictures and text information of multiple products. The picture feature set includes multiple picture features, and the text vector set includes multiple text vectors.

[0072] Optionally, the target product may be a product for which the user is bidding. The real-time feature may represent transaction information of the target product, and the transaction information may include the selling price, previous selling price, and bid information of the seller of the target product.

[0073] Among them, the real-time characteristics of the target product can be determined based on the log message transmitted in real time by the user's bid. The electronic device can calculate the brand price, sale price, bid and other characteristics of the target product in real time based on the log message transmitted in real time by the user's bid to obtain the real-time characteristics of the target product.

[0074] Optionally, the community dynamics information may include community dynamics of all users stored in the electronic device, recording pictures and text information of multiple commodities. The community dynamics may be posts or sales information containing commodities posted by users on e-commerce platforms or other external platforms.

[0075] Optionally, the electronic device may obtain community dynamic information within a preset time period, and create a picture feature set and a text vector set based on the community dynamics.

[0076] Optionally, the image feature set may include multiple image features, and the image features may be products extracted from product images in community dynamics.

[0077] It should be understood that the product images uploaded by users in community dynamics may include content unrelated to the product. For example, when a user takes a product image, content unrelated to the product is mixed into the background. This will undoubtedly interfere with the subsequent judgment of product similarity. Therefore, the products in the product image can be extracted separately as image features.

[0078] In this application, the vgg (Visual Geometry Group) pre-trained model can be used to extract image features of all community dynamics in the e-commerce platform to obtain a set of image features. The dynamic images can be posts in the e-commerce platform or images contained in the sales information of the sold goods. The sales information can be sales posts published by users on the e-commerce platform or product sales pages posted by merchants, etc.

[0079] As another possible implementation, when the product images uploaded by users in the community dynamics only contain products, the images in the bidding log can be directly used as image features.

[0080] Optionally, the text vector set may include text vectors of multiple products. The text vectors may be extracted through a word2vec model from community dynamics on the e-commerce platform, i.e., text information in posts on the platform or sales information of products sold, to finally obtain a text vector set. The word2vec model may also be other word vector generation models, which may be trained based on community content, product descriptions, product titles, etc. on the e-commerce platform.

[0081] S202: extracting the image feature information of the target product from the bidding log, and determining the image similarity feature sequence and the image trend feature sequence of the target product based on the image feature information and the image feature set, wherein the bidding log is used to record the bidding information of the target product, the image similarity feature sequence includes a plurality of numbers arranged in sequence, each number representing the number of image features similar to the image feature information in each time period within a preset time interval, and the image trend feature sequence includes a plurality of trend features arranged in sequence, each trend feature being used to characterize the change information of the number of similar image features in each time period within the preset time interval relative to the previous time period.

[0082] Optionally, the bidding log may be a bidding log of the target product to be identified, including bidding information, pictures, and text information of the target product, wherein the text information may include product description content, product title, etc. of the target product.

[0083] Optionally, the image feature information of the target product may be extracted through a vgg pre-trained model, and the image feature information may be image features of the target product extracted from product images in the bid log.

[0084] It is worth noting that an e-commerce platform may include multiple product images from different angles for a product. The main product image is the most frequently used image and the easiest to identify the product. Therefore, in this application, features of the main image of the target product can be extracted to obtain image feature information of the target product.

[0085] Optionally, the picture similarity feature sequence may include a plurality of numbers arranged in sequence, each number representing the number of picture features similar to the picture feature information in each time period within a preset time interval.

[0086] For example, assuming that it is necessary to perform bid identification based on the trend of changes in commodity popularity in the past month, a picture feature set can be created based on the pictures contained in the community dynamics on the e-commerce platform in the past month, and the similarity between the picture features in the picture feature set and the picture feature information of the target commodity can be calculated. Multiple picture features in multiple picture feature sets that are similar to the picture feature information of the target commodity can be determined, and then these picture features can be arranged in chronological order. The number of picture features similar to the target commodity on each day in the past month can be obtained, that is, the number of posts involving pictures of the target commodity in the community dynamic information. These numbers can be sorted in time to obtain a picture similarity feature sequence.

[0087] Optionally, the image trend feature sequence can describe the popularity trend of the target product within a preset time interval. The image trend feature sequence includes multiple trend features arranged in sequence, and each trend feature is used to characterize the change information of the number of similar image features in each time period within the preset time interval relative to the previous time period.

[0088] The change information can describe the increase in the number of similar image features in the current time period relative to the previous time period, or the decrease in the number of similar image features in the current time period relative to the previous time period. Therefore, the discussion and popularity of the target product on the e-commerce platform within the preset time period can be known through the image trend feature sequence.

[0089] Optionally, the number of similar picture features may be the number in a picture similarity feature sequence, that is, the number of sales information related to the target product in the community content.

[0090] For example, suppose that it is necessary to identify bids based on the commodity popularity change trend in the past month. After obtaining the image similarity feature sequence N[n1,n2,...,n t ], the image trend feature sequence Q[q1,q2,...,q t-1 Specifically, assuming that the preset time interval is one month, n1 represents the number of sales information involving the target product on January 1, and n2 represents the number of sales information involving the target product on January 2. Then, the change information q1 of the number of similar image features on January 2 relative to January 1 can be determined based on n1 and n2.

[0091] S203: extracting the title vector of the target product from the bid log, and determining the text similarity feature sequence and the title trend feature sequence of the target product according to the title vector and the text vector set.

[0092] Optionally, the title vector may be a vectorized representation of the title of the target product and the product description of the target product. The title of the target product and the product description of the target product may be extracted through a word2vec model to obtain a vectorized representation of the title and the product description.

[0093] As a possible implementation, the electronic device can obtain the user's bidding log in real time, perform real-time message transmission and streaming processing on the bidding log, and obtain processed image feature information and text vector information.

[0094] Optionally, the text similarity feature sequence may include multiple numbers arranged in sequence, each number representing the number of text vectors similar to the title vector in each time period within a preset time interval, that is, the number of sales information involving the target product in the community dynamics of the e-commerce platform.

[0095] Optionally, the title trend feature sequence may include a plurality of trend features arranged in sequence, and each trend feature is used to characterize the change information of the number of similar texts in each time period within a preset time interval relative to the previous time period.

[0096] For example, assuming that we need to identify bids based on the popularity trend of a product in the past month, we can determine the community dynamics that are similar to the title vector of the target product in the past month from the text vector set, and arrange the number of these community dynamics in chronological order to obtain the text similarity feature sequence C[c1,c2,...,c t ], and then determine the quantity change information in two adjacent time periods based on the text similarity feature sequence, thereby determining the title trend feature sequence D[d t2 ,d t3 ,...,d t ]. Among them, d t2 It can be determined based on c1 and c2, and is used to describe the change of c2 relative to c1.

[0097] Among them, d t The calculation method of can be shown in the following formula (1), t represents the time period t within the preset time interval T, c t represents the number of texts in the tth time period, c t-1 Indicates the number of texts in the t-1th time period.

[0098] d t =(c t -c t-1 ) / c t-1 (1)

[0099] S204: Inputting the image similarity feature sequence, the image trend feature sequence, the text similarity feature sequence, the title trend feature sequence and the real-time features of the target product into a pre-trained bid recognition model to obtain a bid recognition result of the target product in the bid log.

[0100] Optionally, the bid recognition result may include: abnormal bids and normal bids. Based on the bid recognition result of the target product, the e-commerce platform can determine whether the bid needs to be intercepted through the risk control engine and perform corresponding processing.

[0101] It should be noted that the bid recognition model can output the probability of abnormal bids for goods. By comparing the probability with the preset threshold, the final bid recognition result can be obtained, that is, determining whether the bid is a normal bid or an abnormal bid.

[0102] Optionally, the bid recognition model may be a binary classification model obtained by training a multi-classification tree (Light Gradient Boosting Machine, Light GBM) based on the samples.

[0103] In an embodiment of the present application, by combining image similarity feature sequence, image trend feature sequence, text similarity feature sequence, title trend feature sequence and real-time features of target products for product bid recognition, the popularity trend of the target product in a recent period of time can be more accurately determined through the text information and image information in the community dynamic information, and the impact of product popularity on product price is taken into account. The time series features constructed by multimodal images and texts and structured data are introduced into bid anomaly judgment, which improves the accuracy of identifying abnormal product bids and avoids the problem of low accuracy caused by price prediction based solely on the time series price of the product.

[0104] Next, combine Figure 3 , the step of determining the image similarity feature sequence and the image trend feature sequence of the target product according to the image feature information and the image feature set in the above S202 is described as follows: Figure 3 As shown, the above step S202 includes:

[0105] S301: Determine the similarity between the picture feature information and each picture feature in the picture feature set, and obtain the picture similarity between each picture feature in the picture feature set and the picture feature information.

[0106] Optionally, the similarity between each image feature in the image feature set and the image feature information of the target product may be determined in sequence, and the similarity calculation method may be as shown in the following formula (2).

[0107]

[0108] Wherein, A and B represent image features and image feature information respectively, and n represents the number of image feature information of the target product and the number of image features in the image feature set. It should be understood that the product image of the target product may contain multiple products. In this application, only the similarity between the main image of the target product and each image feature in the image feature set can be calculated, so the number of image feature information of the target product can be 1.

[0109] S302: Determine a picture similarity feature sequence and a picture trend feature sequence based on a preset picture similarity threshold and the picture similarity between each picture feature and the picture feature information.

[0110] Optionally, the image similarity threshold can be used to measure the similarity between the image features and the image feature information. For example, assuming that the image features and the image feature information with similarity between 60% and 100% can be considered to contain the same product, when the similarity between the image features of community dynamics A and the image feature information of the target product is 80%, it can be considered that the target product is included in community dynamics A.

[0111] It is worth noting that the main picture of the product can fully represent the characteristics of the product, and when users make bids, they generally take pictures from the perspective of the main picture of the product. Therefore, if the similarity between the picture in the community content and the main picture of the product meets the preset similarity threshold, it can be determined that the product in the community dynamics and the target product are similar products. This can also ensure that only one picture in a community dynamic will be identified as a picture similar to the target product.

[0112] Next, continue to combine Figure 4 , the step of determining the picture similarity feature sequence and the picture trend feature sequence in the above S302 based on the preset picture similarity threshold and the picture similarity between each picture feature and the picture feature information is described.

[0113] S401: Screening picture features in a picture feature set according to a similarity threshold and a picture similarity between each picture feature and picture feature information to obtain a plurality of target picture features.

[0114] Optionally, based on a similarity threshold, multiple target image features similar to the image feature information of the target product may be screened out from the image feature set, and each target image feature corresponds to a piece of community dynamic information.

[0115] S402: Obtaining a picture similarity feature sequence according to the number of target picture features, the time corresponding to each target picture feature, and a preset time interval.

[0116] It is worth noting that since each target image feature corresponds to a community dynamic, the number of target image features can represent the number of sales information of products related to the target product. When the number of sales information is high, it can be considered that the popularity of the product is high.

[0117] Optionally, each target image feature may include time information. The filtered target time features are first sorted according to the time information, and then the number of target image features belonging to the same time period is counted according to the time periods in the preset time interval. The number of target image features in each time period in the preset time interval can be obtained. The image similarity feature sequence can be obtained by sorting them according to time.

[0118] For example, assuming that the number of image features whose similarity with the target product meets the similarity threshold in the last three days is 12, 25, and 32 respectively, then the image similarity feature sequence is [12, 25, 32].

[0119] As a possible implementation method, the electronic device can also obtain the latest community dynamics on the e-commerce platform in real time, and calculate the similarity between the image in the community dynamics and the image feature information of the target product. If the similarity meets the similarity threshold, the number of the time period of the community dynamics in the image similarity sequence can be increased by 1 according to the time of the community dynamics.

[0120] S403: Perform trend calculation on the picture similarity feature sequence to obtain a picture trend feature sequence.

[0121] Optionally, performing trend calculation on the image similarity feature sequence may be calculating a change in a latter number relative to a former number of two adjacent numbers in the image similarity feature sequence.

[0122] Exemplarily, the calculation method of the image trend feature sequence can be shown in the following formula (3).

[0123] q t =(n t -n t-1 ) / n t-1 (3)

[0124] Among them, q t represents the image trend characteristics of the t-th time period, n t is the number of image features in the tth time period, n t-1 is the number of image features in the t-1th time period.

[0125] In an embodiment of the present application, by calculating the similarity between the pictures of community dynamics in the community content and the pictures of the target product, determining the picture similarity feature sequence, and determining the picture trend feature sequence based on the picture similarity feature sequence, the popularity trend of the target product can be determined through the pictures in the community dynamic information, so that the popularity of the product is also taken into consideration for product bid identification, thereby improving the accuracy of bid identification.

[0126] In some cases, the community updates uploaded by individual users may only include text information. In this application, not only the image popularity trend of the target product can be determined through the images in the community content, but also the text popularity trend of the target product can be determined through the text information in the community content.

[0127] Next, combine Figure 5 , the steps of determining the text similarity feature sequence and the title trend feature sequence of the target product according to the title vector and the text vector set in this application are described, such as Figure 5 As shown, the above step S203 includes:

[0128] S501: Determine the similarity between the title vector and each text vector in the text vector set, and obtain the text similarity between each text vector in the text vector set and the title vector.

[0129] Optionally, the text vectors in the text vector set may include not only the product titles of each community dynamic in the community dynamic information, but also the product descriptions of each community dynamic. For each community dynamic, only a similarity value between the community dynamic and the title vector may be determined.

[0130] S502: Determine a text similarity feature sequence and a title trend feature sequence based on a preset text similarity threshold and the text similarity between each text vector and the title vector.

[0131] Optionally, a text similarity threshold may measure the similarity between the text vector of the community dynamics and the title vector of the target product.

[0132] It is worth noting that each text vector represents a community dynamic, so the number of text vectors can represent the number of community dynamics containing the target product. By counting the text vectors that meet the text similarity threshold according to the time period within the preset time interval, the number of text vectors in each time period can be obtained. By arranging the numbers in time series, the text similarity feature sequence can be obtained.

[0133] Optionally, after determining the text similarity feature sequence, the change information of the latter quantity compared to the previous quantity can be determined according to adjacent quantities in the text similarity feature sequence, and the title trend feature sequence can be obtained by arranging the change information in chronological order.

[0134] It should be noted that the above-mentioned text similarity feature sequence and image similarity feature sequence can be determined through a time window, that is, the time window T is used to determine the text similarity feature sequence and image similarity feature sequence within T consecutive days, and the time window T is continued to be used to determine the change information of the text similarity feature sequence and image similarity feature sequence within T consecutive days, that is, the title trend feature sequence and the image trend feature sequence.

[0135] Next, combine Figure 6 , the steps of determining the similarity between the title vector and each text vector in the text vector set and obtaining the text similarity between each text vector in the text vector set and the title vector are described as follows: Figure 6 As shown, the above step S501 includes:

[0136] S601: Determine the similarity between the word vector of the first word in the title vector and the word vectors of each word in the first text vector, and determine the maximum similarity of each word vector in the first text vector, and use the maximum similarity as the similarity between the first word and the first text vector, the first word is any word in the title vector, and the first text vector is any text vector in the text vector set.

[0137] Optionally, the text vector set includes multiple texts, each of which can represent a community dynamic. In order to determine whether the community dynamic contains a target product, the similarity between the text vector of the community dynamic and the title vector of the target product can be determined.

[0138] S602: Taking the average value of the similarities between each word in the title vector and the first text vector as the similarity between the title vector and the first text vector.

[0139] Taking the similarity calculation between text vector A and title vector as an example, both text vector A and title vector include multiple words. First, for word A in the title vector, the similarity between each word in text vector A and word A is calculated. Then, the largest similarity value in the text vector is used as the similarity of word A in the title vector. The similarity is calculated for each word in the title vector in turn, and finally the average value of the similarities of each word is used as the similarity between the title vector and text vector A.

[0140] It should be noted that in the present application, the above steps S601-S602 can be executed for each text vector in the text vector set to determine the similarity between each text vector in the text vector set and the title vector, and then the above step S502 is executed to determine the text similarity feature sequence and the title trend feature sequence.

[0141] After determining the image similarity feature sequence, image trend feature sequence, text similarity feature sequence and title trend feature sequence, these sequences and the real-time features of the target product can be input into the price recognition model to obtain the bid recognition result of the target product.

[0142] After obtaining the bid recognition result of the target product, the method of the present application further includes:

[0143] The bid of the target product is intercepted or output according to the bid recognition result of the target product.

[0144] As a possible implementation mode, when the bid recognition result of the target commodity indicates that the bid of the target commodity is abnormal and suspected to be speculation, the bid of the target commodity can be intercepted.

[0145] As another possible implementation, when the bid recognition result of the target product indicates that the bid of the target product is normal, the bid can be output normally, that is, the product is published on the platform according to the bid.

[0146] In this application, in order to more comprehensively consider the impact of external public opinion on commodity prices, external public opinion sentiment values ​​and internal community public opinion sentiment values ​​can also be considered as factors for commodity bid identification.

[0147] The external public opinion sentiment value can be determined by monitoring the meaning of words in messages related to the product on external platforms. For example, if most of the messages related to product A on external platforms are derogatory words or words with negative meanings, it can be considered that the external public opinion sentiment value of the product is low.

[0148] The internal community public opinion sentiment value can be determined by monitoring the messages related to the product in the community of the e-commerce platform. When the popularity trend of product A is very high and the internal community public opinion sentiment value is also high, the internal community public opinion sentiment value can also be used as an influencing factor for bid identification and bidding identification and judgment.

[0149] The following is an explanation of the training process of the bid recognition model in this application. Figure 7 As shown, the step includes:

[0150] S701: Obtain sample real-time features of sample target products, and create a sample image feature set and a sample text vector set based on sample community dynamic information.

[0151] Optionally, the sample community dynamics information may include community dynamics of multiple users regarding multiple commodities stored in the electronic device.

[0152] The sample image feature set includes image features of all images in the sample community dynamic information, and the sample text vector set includes text vectors of text information in all sales information in the sample community dynamic information.

[0153] S702: Extracting sample image feature information of the sample target product from the sample bidding log, and determining a sample image similarity feature sequence and a sample image trend feature sequence of the sample target product according to the sample image feature information and the sample image feature set.

[0154] Optionally, each sample product in the sample bidding log can be used as a sample target product for training in turn, and a sample bidding label can be pre-labeled for each bidding log. For example, samples with abnormal bidding can be labeled 1, and samples with normal bidding can be labeled 0.

[0155] S703: extracting a sample title vector of a sample target product from the sample bidding log, and determining a sample text similarity feature sequence and a sample title trend feature sequence of the sample target product according to the sample title vector and the sample text vector set.

[0156] Optionally, the steps of determining the sample image similarity feature sequence, the sample image trend feature sequence, the sample text similarity feature sequence and the sample title trend feature sequence may be the same as the aforementioned steps of determining the image similarity feature sequence, the image trend feature sequence, the text similarity feature sequence and the title trend feature sequence, and this application will not elaborate on them here.

[0157] S704: A price binary classification model is trained using a multi-classification tree according to the sample image similarity feature sequence, the sample image trend feature sequence, the sample text similarity feature sequence, the sample title trend feature sequence and the sample real-time features of the sample target product to obtain a price recognition model.

[0158] Optionally, a binary classification model is trained using a multi-classification tree algorithm for the sample image similarity feature sequence, the sample image trend feature sequence, the sample text similarity feature sequence, the sample title trend feature sequence, and the sample real-time features of the sample products, and a loss value is determined based on the training results and the bid tags of the sample target products. The trained model is iteratively optimized based on the loss value to obtain the final price recognition model.

[0159] Based on the same inventive concept, an abnormal bidding identification device for goods corresponding to the abnormal bidding identification method for goods is also provided in the embodiment of the present application. Since the principle of solving the problem by the device in the embodiment of the present application is similar to the abnormal bidding identification method for goods in the embodiment of the present application, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be repeated.

[0160] Reference Figure 8 FIG. 8 is a schematic diagram of a device for identifying abnormal bidding of a commodity provided in an embodiment of the present application, wherein the device comprises: an acquisition module 801, an image extraction module 802, a text extraction module 803 and an identification module 804, wherein:

[0161] The acquisition module 801 is used to acquire the real-time features of the target product and create a picture feature set and a text vector set according to the community dynamic information. The real-time features are used to characterize the transaction information of the target product. The community dynamic information is used to record the pictures and text information of multiple products. The picture feature set includes multiple picture features, and the text vector set includes multiple text vectors.

[0162] The image extraction module 802 is used to extract the image feature information of the target product from the bidding log, and determine the image similarity feature sequence and the image trend feature sequence of the target product according to the image feature information and the image feature set. The bidding log is used to record the bidding information of the target product. The image similarity feature sequence includes a plurality of sequentially arranged quantities, each of which represents the number of image features similar to the image feature information in each time period within a preset time interval. The image trend feature sequence includes a plurality of sequentially arranged trend features, each of which is used to represent the change information of the number of similar image features in each time period within the preset time interval relative to the previous time period.

[0163] The text extraction module 803 is used to extract the title vector of the target product from the bidding log, and determine the text similarity feature sequence and the title trend feature sequence of the target product according to the title vector and the text vector set;

[0164] The recognition module 804 is used to input the image similarity feature sequence, image trend feature sequence, text similarity feature sequence, title trend feature sequence and real-time features of the target product into the pre-trained bid recognition model to obtain the bid recognition result of the target product in the bid log.

[0165] Optionally, the picture extraction module 802 is specifically used for:

[0166] Determine the similarity between the picture feature information and each picture feature in the picture feature set, and obtain the picture similarity between each picture feature in the picture feature set and the picture feature information;

[0167] Based on a preset picture similarity threshold and the picture similarity between each picture feature and the picture feature information, a picture similarity feature sequence and a picture trend feature sequence are determined.

[0168] Optionally, the picture extraction module 802 is further specifically used for:

[0169] Screening the image features in the image feature set according to a similarity threshold and the image similarity between each image feature and the image feature information to obtain a plurality of target image features;

[0170] According to the number of target image features, the time corresponding to each target image feature and the preset time interval, a picture similarity feature sequence is obtained;

[0171] The trend of the image similarity feature sequence is calculated to obtain the image trend feature sequence.

[0172] Optionally, the text extraction module 803 is specifically used for:

[0173] Determine the similarity between the title vector and each text vector in the text vector set, and obtain the text similarity between each text vector in the text vector set and the title vector;

[0174] Based on a preset text similarity threshold and the text similarity between each text vector and the title vector, a text similarity feature sequence and a title trend feature sequence are determined.

[0175] Optionally, the text extraction module 803 is further specifically used for:

[0176] Determine the similarity between the word vector of the first word in the title vector and the word vectors of each word in the first text vector, and determine the maximum similarity of each word vector in the first text vector, and use the maximum similarity as the similarity between the first word and the first text vector, the first word is any word in the title vector, and the first text vector is any text vector in the text vector set;

[0177] The average value of the similarities between each word in the title vector and the first text vector is taken as the similarity between the title vector and the first text vector.

[0178] Optionally, the abnormal bidding identification device for commodities of the present application may further include a processing module and a training module, wherein the processing module is specifically used for:

[0179] The bid of the target product is intercepted or output according to the bid recognition result of the target product.

[0180] The training module is specifically used for:

[0181] Obtain sample real-time features of sample target products, and create a sample image feature set and a sample text vector set based on sample community dynamic information;

[0182] Extracting sample image feature information of the sample target product from the sample bidding log, and determining a sample image similarity feature sequence and a sample image trend feature sequence of the sample target product based on the sample image feature information and the sample image feature set;

[0183] Extracting a sample title vector of a sample target product from the sample bidding log, and determining a sample text similarity feature sequence and a sample title trend feature sequence of the sample target product based on the sample title vector and the sample text vector set;

[0184] According to the sample image similarity feature sequence, sample image trend feature sequence, sample text similarity feature sequence, sample title trend feature sequence and sample real-time features of sample target products, a multi-classification tree is used to train a price binary classification model to obtain a price recognition model.

[0185] For descriptions of the processing flow of each module in the device and the interaction flow between each module, reference may be made to the relevant descriptions in the above method embodiment, which will not be described in detail here.

[0186] The present application also provides an electronic device, such as Fig. 9 , which is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application, includes: a processor 901, a memory 902 and a bus. The memory 902 stores machine-readable instructions executable by the processor 901 (for example, Figure 8 In the device, the execution instructions corresponding to the acquisition module 801, the image extraction module 802, the text extraction module 803 and the recognition module 804 are obtained, etc. When the computer device is running, the processor 901 communicates with the memory 902 through a bus, and when the machine-readable instructions are executed by the processor 901, the above-mentioned abnormal bidding recognition method for commodities is executed.

[0187] The embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned method for identifying abnormal bidding for commodities are executed.

[0188] Those skilled in the art can clearly understand that, for the convenience and simplicity of description, the specific working process of the system and device described above can refer to the corresponding process in the method embodiment, and will not be repeated in this application. In the several embodiments provided in this application, it should be understood that the disclosed system, device and method can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the modules is only a logical function division. There may be other division methods in actual implementation. For example, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, indirect coupling or communication connection of devices or modules, which can be electrical, mechanical or other forms.

[0189] In addition, each functional unit in each embodiment of the present application can be integrated into a processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or part of the technical solution that contributes to the prior art or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, RandomAccess Memory), disk or optical disk and other media that can store program code.

[0190] The above are only specific implementation methods of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be covered by the protection scope of the present application.

Claims

1. A method for identifying abnormal bidding for commodities, characterized in that: include: Acquire real-time features of a target commodity, and create a picture feature set and a text vector set according to community dynamic information, wherein the real-time features are used to characterize transaction information of the target commodity, and the community dynamic information is used to record pictures and text information of multiple commodities, wherein the picture feature set includes multiple picture features, and the text vector set includes multiple text vectors; Extracting the image feature information of the target product from the bidding log, and determining the image similarity feature sequence and the image trend feature sequence of the target product according to the image feature information and the image feature set, wherein the bidding log is used to record the bidding information of the target product, the image similarity feature sequence includes a plurality of sequentially arranged quantities, each of which represents the number of image features similar to the image feature information in each time period within a preset time interval, and the image trend feature sequence includes a plurality of sequentially arranged trend features, each of which is used to characterize the change information of the number of similar image features in each time period within the preset time interval relative to the previous time period; Extracting the title vector of the target product from the bid log, and determining the text similarity feature sequence and the title trend feature sequence of the target product according to the title vector and the text vector set; The image similarity feature sequence, image trend feature sequence, text similarity feature sequence, title trend feature sequence and real-time features of the target product are input into a pre-trained bid recognition model to obtain a bid recognition result of the target product in the bid log.

2. The method according to claim 1, characterized in that The determining of the image similarity feature sequence and the image trend feature sequence of the target product according to the image feature information and the image feature set includes: Determine the similarity between the picture feature information and each picture feature in the picture feature set, and obtain the picture similarity between each picture feature in the picture feature set and the picture feature information; The picture similarity feature sequence and the picture trend feature sequence are determined based on a preset picture similarity threshold and the picture similarity between each picture feature and the picture feature information.

3. The method according to claim 2, characterized in that The determining the picture similarity feature sequence and the picture trend feature sequence based on a preset picture similarity threshold and the picture similarity between each picture feature and the picture feature information includes: Screening the picture features in the picture feature set according to the similarity threshold and the picture similarity between each picture feature and the picture feature information to obtain a plurality of target picture features; Obtaining the picture similarity feature sequence according to the number of target picture features, the time corresponding to each target picture feature, and the preset time interval; Perform trend calculation on the picture similarity feature sequence to obtain the picture trend feature sequence.

4. The method according to claim 1, characterized in that: The step of determining the text similarity feature sequence and the title trend feature sequence of the target product according to the title vector and the text vector set includes: Determine the similarity between the title vector and each text vector in the text vector set, and obtain the text similarity between each text vector in the text vector set and the title vector; Based on a preset text similarity threshold and the text similarity between each text vector and the title vector, the text similarity feature sequence and the title trend feature sequence are determined.

5. The method according to claim 4, characterized in that The determining the similarity between the title vector and each text vector in the text vector set to obtain the text similarity between each text vector in the text vector set and the title vector includes: Determine the similarity between the word vector of the first word in the title vector and the word vectors of each word in the first text vector, and determine the maximum similarity of each word vector in the first text vector, and use the maximum similarity as the similarity between the first word and the first text vector, the first word is any word in the title vector, and the first text vector is any text vector in the text vector set; The average value of the similarities between each word in the title vector and the first text vector is used as the similarity between the title vector and the first text vector.

6. The method according to claim 1, characterized in that After obtaining the bid recognition result of the target product, the method includes: The bid of the target product is intercepted or outputted according to the bid identification result of the target product.

7. The method according to any one of claims 1 to 6, characterized in that: The step of inputting the picture similarity feature sequence, the picture trend feature sequence, the text similarity feature sequence, the title trend feature sequence and the real-time feature of the target product into a pre-trained price recognition model to obtain the bid recognition result of the target product comprises: Obtain sample real-time features of sample target products, and create a sample image feature set and a sample text vector set based on sample community dynamic information; Extracting sample image feature information of the sample target product from the sample bidding log, and determining a sample image similarity feature sequence and a sample image trend feature sequence of the sample target product according to the sample image feature information and the sample image feature set; Extracting a sample title vector of the sample target product from the sample bidding log, and determining a sample text similarity feature sequence and a sample title trend feature sequence of the sample target product according to the sample title vector and the sample text vector set; The price recognition model is obtained by training a price binary classification model using a multi-classification tree according to the sample image similarity feature sequence, the sample image trend feature sequence, the sample text similarity feature sequence, the sample title trend feature sequence and the sample real-time features of the sample target commodity.

8. A device for identifying abnormal bidding of commodities, characterized in that: include: An acquisition module, used to acquire real-time features of a target commodity, and to create a picture feature set and a text vector set according to community dynamic information, wherein the real-time features are used to characterize transaction information of the target commodity, and the community dynamic information is used to record pictures and text information of multiple commodities, wherein the picture feature set includes multiple picture features, and the text vector set includes multiple text vectors; An image extraction module is used to extract image feature information of a target product from a bidding log, and determine an image similarity feature sequence and an image trend feature sequence of the target product according to the image feature information and the image feature set, wherein the bidding log is used to record the bidding information of the target product, the image similarity feature sequence includes a plurality of sequentially arranged quantities, each of which represents the number of image features similar to the image feature information in each time period within a preset time interval, and the image trend feature sequence includes a plurality of sequentially arranged trend features, each of which is used to characterize change information of the number of similar image features in each time period within the preset time interval relative to the previous time period; A text extraction module, used to extract the title vector of the target product from the bid log, and determine the text similarity feature sequence and the title trend feature sequence of the target product according to the title vector and the text vector set; The recognition module is used to input the image similarity feature sequence, image trend feature sequence, text similarity feature sequence, title trend feature sequence and the real-time features of the target product into a pre-trained bid recognition model to obtain the bid recognition result of the target product in the bid log.

9. An electronic device, characterized in that: include: A processor, a storage medium and a bus, wherein the storage medium stores program instructions executable by the processor. When the electronic device is running, the processor communicates with the storage medium via the bus, and the processor executes the program instructions to perform the steps of the method for identifying abnormal bids for commodities as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method for identifying abnormal bids for commodities according to any one of claims 1 to 7 are executed.

Images