A label verification method, device, equipment and storage medium
By determining the target search keywords of the item label and using the preset semantic matching model for semantic matching, the problem of low matching between item labels and item information in the prior art is solved, and efficient and accurate automatic verification is achieved.
Patent Information
- Application Number
- CN202010239481.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2020-03-30
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2040-03-30
AI Technical Summary
In the prior art, semantic matching between item labels and item information cannot be achieved through character string matching, resulting in low verification accuracy, low manual verification efficiency and increased labor costs.
By determining the target search keywords that match the item label to be verified, the item label and item information are converted into the search keyword and item information matching problem, and a preset semantic matching model trained based on the search exposure click-through rate of sample items is used in advance. Perform semantic matching to measure the degree of match to determine the verification result.
Improve the accuracy and efficiency of label verification, realize automatic verification, reduce manual participation, and improve verification efficiency.
Smart Images

Figure CN111767738B_ABST
Abstract
Description
Technical Field
[0001] The embodiments of the present invention relate to computer technology, and in particular to a label verification method, device, equipment and storage medium. Background Art
[0002] With the rapid development of Internet technology, big data marketing can be carried out by labeling items, fundamentally changing the traditional advertising operation model. Since item labels are actively attached by users when they publish item information, it is necessary to verify the accuracy of item labels to ensure accurate marketing of items.
[0003] Currently, it is usually verified whether the item tag matches the item information by performing string matching between the item tag and the item information, or the backend operator performs manual verification during the review process.
[0004] However, in the process of implementing the present invention, the inventors found that there are at least the following problems in the prior art:
[0005] The string matching method cannot semantically match the item label with the item information. For example, the item information "Korean version of slim fit plus velvet thick warm men's casual trousers sports trousers autumn and winter new fashion slim-fitting trousers business straight-leg suit trousers men's trousers" has a matching degree of 0 with the item label "fashionable fashion", but in fact the two match, which reduces the accuracy of verification. Although the manual verification method can be verified based on the semantics of the item label and item information, the verification efficiency is low and the labor cost is increased. Summary of the invention
[0006] The embodiments of the present invention provide a label verification method, device, equipment and storage medium to realize automatic verification of labels and improve the verification accuracy and verification efficiency of labels.
[0007] In a first aspect, an embodiment of the present invention provides a tag verification method, comprising:
[0008] Obtain target item information and target item labels to be verified;
[0009] Determine a target search keyword that matches the target item tag;
[0010] Inputting the target item information and the target search keyword into a preset semantic matching model, wherein the preset semantic matching model is pre-trained based on the search exposure click rate of sample items;
[0011] The target matching degree between the target item information and the target search keyword is determined according to the output of the preset semantic matching model, and the verification result corresponding to the target item label is determined according to the target matching degree.
[0012] In a second aspect, an embodiment of the present invention further provides a label verification device, comprising:
[0013] A target object label acquisition module is used to acquire target object information and target object labels to be verified;
[0014] A target search keyword determination module, used to determine a target search keyword that matches the target item tag;
[0015] An information input module, used to input the target item information and the target search keyword into a preset semantic matching model, wherein the preset semantic matching model is pre-trained based on the search exposure click rate of sample items;
[0016] The verification result determination module is used to determine the target matching degree between the target item information and the target search keyword according to the output of the preset semantic matching model, and determine the verification result corresponding to the target item label according to the target matching degree.
[0017] In a third aspect, an embodiment of the present invention further provides a device, the device comprising:
[0018] one or more processors;
[0019] A memory for storing one or more programs;
[0020] When the one or more programs are executed by the one or more processors, the one or more processors implement the label verification method as described in any embodiment of the present invention.
[0021] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the label verification method as described in any embodiment of the present invention.
[0022] The embodiments of the above invention have the following advantages or beneficial effects:
[0023] By determining the target search keyword that matches the target item label to be verified, the question of whether the item label matches the item information can be converted into the question of whether the search keyword matches the item information. By using the preset semantic matching model that is pre-trained based on the search exposure click rate of sample items, the target search keyword and the target item information can be semantically matched and the size of the search exposure click rate can be measured at the same time, so that the target matching degree output by the preset semantic matching model can simultaneously reflect the degree of semantic matching and the probability of users clicking on the exposed target item information after searching based on the target search keyword. Therefore, the target matching degree output by the preset semantic matching model can more accurately determine the verification result corresponding to the target item label, thereby improving the accuracy of label verification, and realizing automatic label verification without manual intervention, thereby improving the efficiency of label verification. BRIEF DESCRIPTION OF THE DRAWINGS
[0024] Figure 1 is a flow chart of a label verification method provided by Embodiment 1 of the present invention;
[0025] Figure 2 This is an example of a preset semantic matching model involved in the first embodiment of the present invention;
[0026] Figure 3 This is an example of another preset semantic matching model involved in the first embodiment of the present invention;
[0027] Figure 4 is a flow chart of a label verification method provided by Embodiment 2 of the present invention;
[0028] Figure 5 This is an example of a dictionary tree involved in the second embodiment of the present invention;
[0029] Figure 6 It is a structural schematic diagram of a label verification device provided by Embodiment 3 of the present invention;
[0030] Figure 7 It is a structural schematic diagram of a device provided in Embodiment 4 of the present invention. DETAILED DESCRIPTION
[0031] The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the present invention, rather than to limit the present invention. It should also be noted that, for ease of description, only parts related to the present invention, rather than all structures, are shown in the accompanying drawings.
[0032] Embodiment 1
[0033] Figure 1This is a flowchart of a tag verification method provided in the first embodiment of the present invention. This embodiment can be applied to the case of verifying whether the item tag matches the item information, especially the scenario of verifying the tags attached to items in the e-commerce platform. The method can be executed by a tag verification device, which can be implemented by software and / or hardware and integrated in a device with data functions. Figure 1 As shown, the method specifically comprises the following steps:
[0034] S110: Obtain target object information and target object label to be verified.
[0035] Among them, the item tag is a word that is generated in advance based on the item information and has the nature of a shopping guide and is aggregated according to marketing logic, such as "fashion trend", "Valentine's Day", "One Piece", etc. The target item tag may refer to the item tag to be verified to which the target item is attached. The target item information may refer to the specific feature information of the target item. For example, the target item information may include but is not limited to: at least one of the target item title information, the target item attribute information and the target item category information. Among them, the target item title information may be brief information for introducing the target item. The target item attribute information may be detailed information for introducing the target item, such as the target item attribute information may include the target item model, the target item is suitable for the crowd, etc. The target item category information may refer to the information of the category to which the target item belongs, for example, if the target item is a sweater, then its corresponding category information is tops.
[0036] Specifically, when verifying whether the object tag matches the object information, the target object information corresponding to the target object and the target object tag to be verified may be obtained.
[0037] S120: Determine a target search keyword that matches the target item tag.
[0038] The search keyword may refer to the query search information entered by the user when searching for an item. The target search keyword may refer to the search keyword that matches the target item label. For example, if the target item label is "retro style" and a search keyword is "men's retro style", then it can be determined that the target search keyword that matches the target item label "retro style" is "men's retro style". In this embodiment, the number of target search keywords that match the target item label can be one or more.
[0039] Specifically, each search keyword in the search database can be matched with the target item tag based on a preset matching method, and the successfully matched search keyword can be determined as the target search keyword. Among them, the preset matching method can be but not limited to the edit distance matching method and the vector cosine distance matching method. For example, the matching similarity between each search keyword in the search database and the target item tag can be determined based on the preset matching method, and each search keyword with a matching similarity greater than a preset similarity can be determined as the target search keyword; each search keyword can also be arranged in descending order based on the matching similarity, and the first preset number of search keywords after the arrangement can be determined as the target search keyword. This embodiment can convert the problem of whether the item tag matches the item information into the problem of whether the search keyword matches the item information by determining the target search keyword that matches the target item tag, so as to verify whether the item tag matches the item information by verifying the search keyword and the item information.
[0040] S130: Input the target item information and the target search keyword into a preset semantic matching model, wherein the preset semantic matching model is pre-trained based on the search exposure click rate of the sample item.
[0041] Among them, the search exposure click rate may refer to the ratio of the number of times an item is clicked and the number of times it is exposed after a user searches based on a search keyword. The number of times exposed may refer to the total number of times item information is displayed after a search. The number of clicks may refer to the total number of times a user clicks on the displayed item information after a search. The sample items may be items that are pre-screened based on the search exposure click rate of the items and used to train a preset semantic matching model. The preset semantic matching model may be a pre-set network model for calculating the semantic similarity between item information and search keywords. The preset semantic matching model is obtained by pre-training with the search exposure click rate as a learning target, so that the output result of the preset semantic matching model can simultaneously reflect the semantic similarity and the size of the user's post-search exposure click rate, so that the item information can be verified based on the semantic similarity and the user's click behavior to see whether the search keyword matches the item information, which greatly improves the accuracy of the verification. Exemplarily, the preset semantic matching model may be, but is not limited to: DSSM (Deep Structured Semantic Models) deep structured semantic model and CTR (Click-Through Rate) click-through rate prediction model. The DSSM deep structured semantic model may be, but is not limited to, a DSSM deep structured semantic model based on an LSTM (Long-Short-Term Memory) long short-term memory network model, namely, LSTM-DSSM. The CTR click rate prediction model may be, but is not limited to, a FTRL (Follow The Regularized Leader) online optimization model.
[0042] Specifically, the target item information and the target search keywords are input into a pre-trained preset semantic matching model, so that the preset semantic matching model can perform semantic matching on the target item information and the target search keywords, and at the same time analyze the click probability of the user on the exposed target item information after searching based on the target search keywords, thereby obtaining the target matching degree between the target item information and the target search keywords, and outputting it.
[0043] S140: Determine a target matching degree between the target item information and the target search keyword according to an output of a preset semantic matching model, and determine a verification result corresponding to the target item tag according to the target matching degree.
[0044] Among them, the target matching degree may refer to the matching similarity between the target item information and the target search keyword, which may simultaneously reflect the semantic matching degree and the probability of users clicking on the exposed target item information after searching based on the target search keyword. For example, if the target matching degree is higher, it indicates that the semantic matching degree and the exposure click rate of the target item information after searching based on the target search keyword are both higher, that is, the target item information is more matched with the target search keyword. The verification result corresponding to the target item tag may refer to the result of whether the target item tag matches the target item information, which may include two results: verification success and verification failure.
[0045] Specifically, the target matching degree between the target item information and the target search keyword can be obtained according to the output of the preset semantic matching model. Since the target search keyword matches the target item label, the target matching degree can also be used to characterize the matching degree between the target item information and the target item label, so that the item information and the item label can be verified based on the semantic similarity and the user's click behavior, which greatly improves the accuracy of the label verification, and the entire verification process does not require manual participation, realizes the automatic verification of the label, and improves the efficiency of the label verification. This embodiment can determine the verification result corresponding to the target item label by comparing whether the target matching degree is greater than the preset matching degree. For example, when the target matching degree is greater than the preset matching degree, the verification result corresponding to the target item label can be determined as: the target item information matches the target item label, that is, the verification is successful, otherwise it is determined that the target item information does not match the target item label, that is, the verification fails.
[0046] It should be noted that the preset semantic matching model can also determine the verification result corresponding to the target object label based on the target matching degree between the target object information and the target search keyword and the pre-trained preset matching degree, and output the verification result. Therefore, the verification result corresponding to the target object label can be directly obtained based on the output of the preset semantic matching model, thereby further improving the label verification efficiency.
[0047] The technical solution of this embodiment can convert the problem of whether the item label matches the item information into the problem of whether the search keyword matches the item information by determining the target search keyword that matches the target item label to be verified. By using the preset semantic matching model obtained by pre-training based on the search exposure click rate of sample items, the target search keyword and the target item information can be semantically matched and the size of the search exposure click rate can be measured at the same time, so that the target matching degree output by the preset semantic matching model can simultaneously reflect the degree of semantic matching and the probability of users clicking on the exposed target item information after searching based on the target search keyword, so that the target matching degree output by the preset semantic matching model can more accurately determine the verification result corresponding to the target item label, improve the accuracy of label verification, and realize automatic label verification without manual participation, thereby improving the efficiency of label verification.
[0048] Based on the above technical solutions, Figure 2 An example of a preset semantic matching model is given. Figure 2 As shown, the preset semantic matching model may include: a word vector sequence determination sub-model, a preset recurrent network sub-model and a matching degree determination sub-model.
[0049] Among them, the word vector sequence determination submodel is used to: perform word segmentation processing on the input target item information and target search keywords respectively, and determine the first word vector sequence corresponding to the target item information and the second word vector sequence corresponding to the target search keywords, and input the first word vector sequence and the second word vector sequence into the preset recurrent network submodel. Among them, the word vector sequence determination submodel can refer to language processing of text in order to obtain data of the input type of the preset recurrent network submodel, that is, to obtain the first word vector sequence and the second word vector sequence. For example, the word vector sequence determination submodel can be but is not limited to the VSM (Vector Space Model) vector space model, the Word2vec model and the BERT (Bidirectional Encoder Representation from Transformers) bidirectional encoder model. The target item information and target search keywords in this embodiment may contain a large number of long-tail words, such as a large number of numbers, English combinations of item numbers, item models and item capacities, etc., which makes it impossible to exhaust the word segmentation and will lose useful information. Therefore, when the target item information and the target search keywords are segmented, each word in the target item information and the target search keywords can be treated as a word to obtain the first word vector sequence and the second word vector sequence under the "word granularity" to further improve the verification accuracy.
[0050] The preset recurrent network sub-model is used to: perform vector integration on the input first word vector sequence and the second word vector sequence, determine the first text vector corresponding to the target item information and the second text vector corresponding to the target search keyword, and input the first text vector and the second text vector into the matching degree determination sub-model. The preset recurrent network sub-model may be a network model that performs vector integration on the word vector sequence based on deep learning. For example, the preset recurrent network sub-model may be, but is not limited to, an LSTM long short-term memory network model.
[0051] The matching degree determination submodel is used to: perform vector matching on the input first text vector and the second text vector, determine the target matching degree between the target item information and the target search keyword, and output the target matching degree. The matching degree determination submodel may be a model that determines the target matching degree between the target item information and the target search keyword based on a preset similarity calculation method. The preset similarity calculation method may be, but is not limited to, a vector cosine distance method.
[0052] Based on the above technical solutions, Figure 3 Another example of a preset semantic matching model is given. Figure 3 As shown, the preset semantic matching model may include, in addition to the above-mentioned word vector sequence determination sub-model, preset recurrent network sub-model and matching degree determination sub-model, a self-attention mechanism sub-model.
[0053] Among them, the word vector sequence determination sub-model is also used to: input the first word vector sequence and the second word vector sequence into the self-attention mechanism sub-model, so that the self-attention mechanism sub-model can obtain the first word vector sequence and the second word vector sequence.
[0054] Among them, the self-attention mechanism sub-model is used to: calculate the attention weights of the input first word vector sequence and the second word vector sequence, determine the first attention weight corresponding to each first word vector in the first word vector sequence, and the second attention weight corresponding to each second word vector in the second word vector sequence, and input each first attention weight and each second attention weight into the preset recurrent network sub-model. Among them, the self-attention mechanism sub-model can be a model that weights each word vector based on the Self-attention mechanism to increase the attention of useful information and reduce the attention of useless information. The first attention weight can be used to characterize the importance of each word in the target item information for information matching. The second attention weight can be used to characterize the importance of each word in the target search keyword for information matching.
[0055] Among them, the preset recurrent network sub-model is specifically used to: based on the input first attention weights and the second attention weights, perform vector integration on the input first word vector sequence and the second word vector sequence, determine the first text vector corresponding to the target item information and the second text vector corresponding to the target search keyword, and input the first text vector and the second text vector into the matching degree determination sub-model. Specifically, the preset recurrent network sub-model can perform weighted processing on the corresponding first word vector based on each first attention weight, so that the important first word vector in the target item information increases and the unimportant first word vector decreases, and perform weighted processing on the corresponding second word vector based on each second attention weight, so that the important second word vector in the target search keyword increases and the unimportant second word vector decreases, so that according to the weighted first word vector sequence and the second word vector sequence, a more representative first text vector and a second text vector can be generated, thereby improving the accuracy of matching.
[0056] On the basis of the above technical solution, before using the preset semantic matching model, it is necessary to train the preset semantic matching model based on the search exposure click rate of the sample items. For example, the training process of the preset semantic matching model may include: based on the search exposure click rate of the items, sampling each item in the item database to obtain multiple sample data, the sample data including sample item information, sample search keywords and standard matching degree; inputting each sample data into the preset semantic matching model, determining the output matching degree corresponding to each sample data according to the output of the preset semantic matching model; determining the training error according to the output matching degree and the corresponding standard matching degree; taking the training error reaching the convergence condition as the training target, and training the preset semantic matching model.
[0057] Among them, the sample search keyword may refer to a search keyword that matches the sample item label of the sample item. The standard matching degree may be used to characterize the verification result corresponding to the sample item label. For example, if the sample item label matches the sample item information, the standard matching degree is determined to be 1. If the sample item label does not match the sample item information, the standard matching degree is determined to be 0. The standard matching degree in this embodiment may be determined based on the search exposure click rate of the sample item. The sample data may include positive sample data and negative sample data, wherein the standard matching degree in the positive sample data may be 1, and the standard matching degree in the negative sample data may be 0.
[0058] Specifically, based on the search exposure click rate of each item in the item database, sample items for model training can be screened out from each item in the item database, and sample data corresponding to each sample item can be obtained. The sample item information and sample search keywords in each sample data can be input into the preset semantic matching model to be trained, and the output matching degree output by the preset semantic matching model can be obtained. Based on the loss function, the training error can be determined according to the output matching degree and the corresponding standard matching degree, for example, the difference between the output matching degree and the corresponding standard matching degree can be determined as the training error. Based on the training error, it can be detected whether the convergence condition is currently reached, such as whether the training error is less than the preset error or whether the training error change tends to be stable, or whether the current number of iterations is equal to the preset number. If it is detected that the convergence condition is currently reached, such as the training error is less than the preset error or the error change tends to be stable, or the current number of iterations is equal to the preset number, it indicates that the training of the preset semantic matching model is completed, and the iterative training can be stopped at this time. If it is detected that the convergence condition is not currently reached, the training error can be back-propagated to the preset semantic matching model, and the network parameters in the preset semantic matching model are adjusted until the convergence condition is reached.
[0059] Exemplarily, based on the search exposure click rate of the items, sampling each item in the item database to obtain multiple sample data can include: determining the search exposure click rate of each item in the item database; determining the item data with a search exposure click rate greater than a first preset threshold as positive sample data, and determining the standard matching degree corresponding to the positive sample data to be 1; determining the item data with a search exposure click rate less than a second preset threshold as negative sample data, and determining the standard matching degree corresponding to the negative sample data to be 0.
[0060] The first preset threshold may be a pre-set value higher than the average search exposure click rate, used to represent the minimum search exposure click rate when the search keyword matches the item information. The second preset threshold may be a pre-set value lower than the average search exposure click rate, used to represent the maximum search exposure click rate when the search keyword does not match the item information.
[0061] Specifically, based on the search log of each item in the item database, the number of times each item is exposed and clicked can be counted, and the ratio of the number of clicks to the number of exposures can be used as the search exposure click rate of the corresponding item. If the search exposure click rate of an item is greater than the first preset threshold, the item data of the item is determined as positive sample data, and the standard matching degree in the positive sample data is determined to be 1. If the search exposure click rate of an item is less than the second preset threshold, the item data of the item is determined as negative sample data, and the standard matching degree in the negative sample data is determined to be 0, so that the preset semantic matching model can be trained based on the search exposure click rate of the sample item.
[0062] It should be noted that, when determining negative sample data, this embodiment may also randomly sample a portion of item data as negative sample data after searching based on search keywords, and eliminate item data whose search exposure click-through rate is greater than a first preset threshold, so as to avoid using item data whose search exposure click-through rate is greater than the first preset threshold as negative sample data, thereby ensuring the globality of sample distribution, improving the training effect of the preset semantic matching model, and further improving the accuracy of label verification.
[0063] Embodiment 2
[0064] Figure 4 This is a flowchart of a tag verification method provided in Embodiment 2 of the present invention. Based on the above embodiment, this embodiment optimizes the step of "determining a target search keyword that matches the target item tag". The explanations of the terms that are the same or corresponding to the above embodiment are not repeated here.
[0065] See also Figure 2 The tag verification method provided in this embodiment specifically includes the following steps:
[0066] S210: Obtain target object information and target object label to be verified.
[0067] S220: Expand similar tags for the target item tag to determine a target similar tag corresponding to the target item tag.
[0068] The target similarity tag may refer to a tag that is semantically similar to the target item tag, such as a synonym and / or near synonym of the target item tag. For example, if the target item tag is "retro style", the target similarity tag may be "retro style", "nostalgic style" and "fashion retro" etc.
[0069] Specifically, the target item label is usually composed of fewer characters, such as 2 to 5 Chinese characters, so that the target item label has less information. Therefore, it is necessary to increase the information content of the target item label by expanding the target item label with similar labels to further improve the accuracy of label verification.
[0070] Exemplarily, S220 may include: obtaining the word segmentation results of each item information in the item database and each search keyword in the search database; determining each similar word that co-occurs with the target item tag and the number of co-occurrences corresponding to each similar word based on the word segmentation results; and determining the target similar tag corresponding to the target item tag from each similar word based on the number of co-occurrences.
[0071] Specifically, based on the word segmentation dictionary, each item information in the item database and each search keyword in the search database can be segmented to obtain the word segmentation results. Each word in the analysis result is matched with the target item label, and the number of co-occurrences corresponding to each similar word that appears in the same item information or the same search keyword as the target item label is counted. Each similar word whose number of co-occurrences is greater than a preset number can be determined as a target similarity label; each similar word can also be arranged in descending order based on the number of co-occurrences, and the first preset number of similar words after arrangement can be determined as the target similarity label, thereby realizing the similarity label expansion of the target item label. This embodiment can also manually review and screen the target similarity label determined based on the number of co-occurrences, and use the screened target similarity label as the final target similarity label to further improve the accuracy of the target similarity label determination.
[0072] S230, matching the target item tag with each search keyword in the search database to determine a first search keyword that matches the target item tag, and matching the target similarity tag with each search keyword in the search database to determine a second search keyword that matches the target similarity tag.
[0073] Specifically, each search keyword in the search database can be matched with the target item tag based on matching methods such as edit distance or vector cosine distance, and the successfully matched search keyword can be determined as the first search keyword, and each search keyword in the search database can be matched with the target similar tag, and the successfully matched search keyword can be determined as the second search keyword.
[0074] Exemplarily, S230 may include: obtaining a dictionary tree created based on each item tag in the tag database, and a failure pointer corresponding to each node in the dictionary tree, wherein the failure pointer is used to jump to the pointed node when the character matching of the corresponding node fails; based on the dictionary tree and the failure pointer in the dictionary tree, character matching is performed on each search keyword in the search database with each item tag to determine the matching item tag corresponding to each search keyword; based on the matching item tag corresponding to each search keyword, a first search keyword matching the target item tag and a second search keyword matching the target similar tag are determined.
[0075] Specifically, the target item label and the target similarity label can be quickly aligned with the search keyword based on the multi-mode matching method of AC (Aho-Corasick) automaton, thereby improving the matching efficiency of the target item label and the target similarity label. Figure 5 An example of a dictionary tree is given. Figure 5 As shown, this embodiment can create a dictionary tree Trie based on each item tag in the tag database, and establish a failure pointer for each node in the dictionary tree, so that when the character matching of the node fails, it jumps to the node pointed to by the failure pointer of the node to continue matching, thereby using the common prefix of the string to reduce the query time and minimize the comparison of useless strings, which has higher query efficiency than the hash tree. By using the failure pointer in the dictionary tree, each search keyword in the search database can be matched with characters in the dictionary tree, and the matching item tag that matches each search keyword can be quickly determined, so that based on the matching item tag corresponding to each search keyword, the first search keyword that matches the target item tag and the second search keyword that matches the target similar tag can be determined. For example, Figure 5 As shown in the figure, if the search keyword is "Men's Retro Style", the matching item label corresponding to the search keyword can be determined in the dictionary tree as "Retro Style". If the search keyword is "Fashionable Casual Style", the matching item label corresponding to the search keyword can be determined in the dictionary tree as "Casual Style". When the target item label is "Casual Style", it can be determined that the first search keyword matching the target item label is "Fashionable Casual Style", thereby improving the alignment efficiency of item labels and search keywords.
[0076] S240: Determine the first search keyword and the second search keyword as target search keywords.
[0077] Specifically, by determining both the first search keyword and the second search keyword as target search keywords, the amount of information of the target search keywords can be expanded so that the preset semantic matching model can semantically match multiple target search keywords with target item information, further improving the matching accuracy, and thereby also improving the accuracy of label verification.
[0078] S250: Input the target item information and the target search keyword into a preset semantic matching model, wherein the preset semantic matching model is pre-trained based on the search exposure click rate of the sample item.
[0079] S260: Determine a target matching degree between the target item information and the target search keyword according to an output of a preset semantic matching model, and determine a verification result corresponding to the target item tag according to the target matching degree.
[0080] The technical solution of this embodiment determines the target similarity tag corresponding to the target object tag by expanding the similarity tag of the target object tag, and determines the first search keyword matching the target object tag and the second search keyword matching the target similarity tag as the target search keywords, thereby expanding the information volume of the target search keywords, so that the preset semantic matching model can semantically match multiple target search keywords with the target object information, further improving the matching accuracy, and thus also improving the accuracy of label verification.
[0081] The following is an embodiment of a label verification device provided by an embodiment of the present invention. The device and the label verification methods of the above-mentioned embodiments belong to the same inventive concept. For details not described in detail in the embodiment of the label verification device, please refer to the embodiment of the above-mentioned label verification method.
[0082] Embodiment 3
[0083] Figure 6 This is a structural schematic diagram of a label verification device provided in Example 3 of the present invention. This embodiment can be applied to the situation of verifying whether the item label and item information match. The device specifically includes: a target item label acquisition module 310, a target search keyword determination module 320, an information input module 330 and a verification result determination module 340.
[0084] Among them, the target item label acquisition module 310 is used to obtain the target item information and the target item label to be verified; the target search keyword determination module 320 is used to determine the target search keyword that matches the target item label; the information input module 330 is used to input the target item information and the target search keyword into a preset semantic matching model, wherein the preset semantic matching model is pre-trained based on the search exposure click rate of the sample item; the verification result determination module 340 is used to determine the target matching degree between the target item information and the target search keyword according to the output of the preset semantic matching model, and determine the verification result corresponding to the target item label according to the target matching degree.
[0085] The technical solution of this embodiment can convert the problem of whether the item label matches the item information into the problem of whether the search keyword matches the item information by determining the target search keyword that matches the target item label to be verified. By using the preset semantic matching model obtained by pre-training based on the search exposure click rate of sample items, the target search keyword and the target item information can be semantically matched and the size of the search exposure click rate can be measured at the same time, so that the target matching degree output by the preset semantic matching model can simultaneously reflect the degree of semantic matching and the probability of users clicking on the exposed target item information after searching based on the target search keyword, so that the target matching degree output by the preset semantic matching model can more accurately determine the verification result corresponding to the target item label, improve the accuracy of label verification, and realize automatic label verification without manual participation, thereby improving the efficiency of label verification.
[0086] Optionally, the target search keyword determination module 320 includes:
[0087] A target similarity tag determination unit, used to perform similarity tag expansion on the target object tag and determine the target similarity tag corresponding to the target object tag;
[0088] A search keyword matching unit, used to match the target item tag with each search keyword in the search database, determine a first search keyword that matches the target item tag, and match the target similarity tag with each search keyword in the search database, determine a second search keyword that matches the target similarity tag;
[0089] The target search keyword determination unit is used to determine the first search keyword and the second search keyword as target search keywords.
[0090] Optionally, the target similarity label determination unit is specifically configured to:
[0091] Obtain the word segmentation results of each item information in the item database and each search keyword in the search database; determine each similar word that co-occurs with the target item label and the number of co-occurrences corresponding to each similar word based on the word segmentation results; determine the target similar label corresponding to the target item label from each similar word based on the number of co-occurrences.
[0092] Optionally, the search keyword matching unit is specifically used to:
[0093] Obtain a dictionary tree created according to each item tag in the tag database, and a failure pointer corresponding to each node in the dictionary tree, wherein the failure pointer is used to jump to the node pointed to when the character matching of the corresponding node fails; based on the dictionary tree and the failure pointer in the dictionary tree, perform character matching on each search keyword in the search database with each item tag to determine the matching item tag corresponding to each search keyword; according to the matching item tag corresponding to each search keyword, determine a first search keyword that matches the target item tag, and a second search keyword that matches the target similar tag.
[0094] Optionally, the preset semantic matching model includes: a word vector sequence determination sub-model, a preset recurrent network sub-model and a matching degree determination sub-model; wherein,
[0095] The word vector sequence determination sub-model is used to: perform word segmentation processing on the input target item information and the target search keyword respectively, and determine the first word vector sequence corresponding to the target item information and the second word vector sequence corresponding to the target search keyword, and input the first word vector sequence and the second word vector sequence into the preset recurrent network sub-model;
[0096] The preset recurrent network sub-model is used to: perform vector integration on the input first word vector sequence and the second word vector sequence, determine the first text vector corresponding to the target item information and the second text vector corresponding to the target search keyword, and input the first text vector and the second text vector into the matching degree determination sub-model;
[0097] The matching degree determination sub-model is used to: perform vector matching on the input first text vector and the second text vector, determine the target matching degree between the target item information and the target search keyword, and output the target matching degree.
[0098] Optionally, the preset semantic matching model further includes: a self-attention mechanism sub-model;
[0099] Accordingly, the word vector sequence determination sub-model is further used to: input the first word vector sequence and the second word vector sequence into the self-attention mechanism sub-model;
[0100] The self-attention mechanism sub-model is used to: perform attention weight calculation on the input first word vector sequence and the second word vector sequence, determine the first attention weight corresponding to each first word vector in the first word vector sequence, and the second attention weight corresponding to each second word vector in the second word vector sequence, and input each first attention weight and each second attention weight into the preset recurrent network sub-model;
[0101] The preset recurrent network sub-model is specifically used to: based on the input first attention weights and the second attention weights, perform vector integration on the input first word vector sequence and the second word vector sequence, determine the first text vector corresponding to the target item information and the second text vector corresponding to the target search keyword, and input the first text vector and the second text vector into the matching degree determination sub-model.
[0102] Optionally, the device further comprises:
[0103] The preset semantic matching model training model is used to: sample each item in the item database based on the search exposure click rate of the item to obtain multiple sample data, the sample data including sample item information, sample search keywords and standard matching degree; input each sample data into the preset semantic matching model, and determine the output matching degree corresponding to each sample data according to the output of the preset semantic matching model; determine the training error according to the output matching degree and the corresponding standard matching degree; take the training error reaching the convergence condition as the training target, and train the preset semantic matching model.
[0104] Optionally, the preset semantic matching model training model is also used to: determine the search exposure click-through rate of each item in the item database; determine the item data whose search exposure click-through rate is greater than a first preset threshold as positive sample data, and determine the standard matching degree corresponding to the positive sample data to be 1; determine the item data whose search exposure click-through rate is less than a second preset threshold as negative sample data, and determine the standard matching degree corresponding to the negative sample data to be 0.
[0105] Optionally, the target item information includes at least one of target item title information, target item attribute information and target item category information.
[0106] The label verification device provided in the embodiment of the present invention can execute the label verification method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects for executing the label verification method.
[0107] It is worth noting that in the embodiment of the above-mentioned label verification device, the various units and modules included are only divided according to functional logic, but are not limited to the above-mentioned division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention.
[0108] Embodiment 4
[0109] Figure 7 A schematic diagram of the structure of a device provided in Embodiment 4 of the present invention. Figure 7 A block diagram of an exemplary device 12 suitable for use in implementing embodiments of the present invention is shown. Figure 7 The device 12 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.
[0110] like Figure 7 As shown, device 12 is in the form of a general purpose computing device. Components of device 12 may include, but are not limited to, one or more processors or processing units 16, system memory 28, and bus 18 that connects various system components (including system memory 28 and processing unit 16).
[0111] Bus 18 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processor or a local bus using any of a variety of bus architectures. By way of example, these architectures include, but are not limited to, an Industry Standard Architecture (ISA) bus, a Micro Channel Architecture (MAC) bus, an Enhanced ISA bus, a Video Electronics Standards Association (VESA) local bus, and a Peripheral Component Interconnect (PCI) bus.
[0112] Device 12 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by device 12, including volatile and non-volatile media, removable and non-removable media.
[0113] System memory 28 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and / or cache memory 32. Device 12 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, storage system 34 may be used to read and write non-removable, non-volatile magnetic media ( Figure 7 not shown, usually called a "hard drive"). Although Figure 7Not shown in the figure, a disk drive for reading and writing to a removable non-volatile disk (e.g., a "floppy disk"), and an optical disk drive for reading and writing to a removable non-volatile optical disk (e.g., a CD-ROM, a DVD-ROM, or other optical media) may be provided. In these cases, each drive may be connected to the bus 18 via one or more data medium interfaces. The system memory 28 may include at least one program product having a set (e.g., at least one) of program modules that are configured to perform the functions of the various embodiments of the present invention.
[0114] A program / utility 40 having a set (at least one) of program modules 42 may be stored, for example, in system memory 28, such program modules 42 including, but not limited to, an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment. Program modules 42 generally perform the functions and / or methods of the embodiments described herein.
[0115] The device 12 may also communicate with one or more external devices 14 (e.g., keyboards, pointing devices, displays 24, etc.), may communicate with one or more devices that enable a user to interact with the device 12, and / or may communicate with any device that enables the device 12 to communicate with one or more other computing devices (e.g., network cards, modems, etc.). Such communication may be performed via an input / output (I / O) interface 22. Furthermore, the device 12 may also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via a network adapter 20. As shown, the network adapter 20 communicates with other modules of the device 12 via a bus 18. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the device 12, including, but not limited to, microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.
[0116] The processing unit 16 executes various functional applications and data processing by running the program stored in the system memory 28, for example, implementing a tag verification method provided by the embodiment of the present invention, the method comprising:
[0117] Obtain target item information and target item labels to be verified;
[0118] Determine target search keywords that match target item tags;
[0119] Inputting target item information and target search keywords into a preset semantic matching model, wherein the preset semantic matching model is pre-trained based on the search exposure click rate of sample items;
[0120] The target matching degree between the target item information and the target search keyword is determined according to the output of the preset semantic matching model, and the verification result corresponding to the target item label is determined according to the target matching degree.
[0121] Of course, those skilled in the art can understand that the processor can also implement the technical solution of the method for determining the reserved inventory amount provided in any embodiment of the present invention.
[0122] Embodiment 5
[0123] This embodiment provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, a tag verification method according to any embodiment of the present invention is implemented. The method includes:
[0124] Obtain target item information and target item labels to be verified;
[0125] Determine target search keywords that match target item tags;
[0126] Inputting target item information and target search keywords into a preset semantic matching model, wherein the preset semantic matching model is pre-trained based on the search exposure click rate of sample items;
[0127] The target matching degree between the target item information and the target search keyword is determined according to the output of the preset semantic matching model, and the verification result corresponding to the target item label is determined according to the target matching degree.
[0128] The computer storage medium of the embodiment of the present invention can adopt any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to: an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, a computer-readable storage medium can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it.
[0129] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, which carry computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.
[0130] The program code embodied on the computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.
[0131] Computer program code for performing the operations of the present invention may be written in one or more programming languages or a combination thereof, including object-oriented programming languages such as Java, Smalltalk, C++, and conventional procedural programming languages such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate 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 through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).
[0132] It should be understood by those skilled in the art that the modules or steps of the present invention described above can be implemented by a general-purpose computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, optionally, they can be implemented by a program code executable by a computer device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.
[0133] Note that the above are only preferred embodiments of the present invention and the technical principles used. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in more detail through the above embodiments, the present invention is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present invention, and the scope of the present invention is determined by the scope of the appended claims.
Claims
1. A label verification method, characterized in that: include: Obtain target item information and target item labels to be verified; Determine a target search keyword that matches the target item tag; Inputting the target item information and the target search keyword into a preset semantic matching model, wherein the preset semantic matching model is pre-trained based on sample item information, sample search keywords and a standard matching degree, and the standard matching degree is determined based on the search exposure click rate of the sample item; Determine the target matching degree between the target item information and the target search keyword according to the output of the preset semantic matching model, and determine the verification result corresponding to the target item label according to the target matching degree, wherein the verification result refers to the result of whether the target item label matches the target item information; Among them, the training process of the preset semantic matching model includes: based on the search exposure click rate of the items, sampling each item in the item database to obtain multiple sample data, the sample data including sample item information, sample search keywords and standard matching degree; inputting each of the sample data into the preset semantic matching model, and determining the output matching degree corresponding to each of the sample data according to the output of the preset semantic matching model; determining the training error according to the output matching degree and the corresponding standard matching degree; taking the training error reaching the convergence condition as the training target, and training the preset semantic matching model.
2. The method according to claim 1, characterized in that Determining a target search keyword that matches the target item tag includes: Expanding similar tags for the target item tag to determine a target similar tag corresponding to the target item tag; Matching the target item tag with each search keyword in the search database to determine a first search keyword that matches the target item tag, and matching the target similarity tag with each search keyword in the search database to determine a second search keyword that matches the target similarity tag; The first search keyword and the second search keyword are determined as target search keywords.
3. The method according to claim 2, characterized in that Expanding the target item tag with similar tags to determine a target similar tag corresponding to the target item tag includes: Obtaining information about each item in the item database and segmentation results of each search keyword in the search database; According to the word segmentation result, determine each similar word that co-occurs with the target item label and the number of co-occurrences corresponding to each similar word; According to each of the co-occurrence times, a target similar tag corresponding to the target object tag is determined from each of the similar words.
4. The method according to claim 2, characterized in that: Matching the target item tag with each search keyword in a search database to determine a first search keyword that matches the target item tag, and matching the target similarity tag with each search keyword in a search database to determine a second search keyword that matches the target similarity tag, including: Obtain a dictionary tree created according to each item tag in the tag database, and a failure pointer corresponding to each node in the dictionary tree, wherein the failure pointer is used to jump to the node pointed to when the character matching of the corresponding node fails; Based on the dictionary tree and the failure pointer in the dictionary tree, character matching is performed between each search keyword in the search database and each item tag to determine a matching item tag corresponding to each search keyword; According to the matching object tag corresponding to each search keyword, a first search keyword matching the target object tag and a second search keyword matching the target similar tag are determined.
5. The method according to claim 1, characterized in that The preset semantic matching model includes: a word vector sequence determination sub-model, a preset recurrent network sub-model and a matching degree determination sub-model; wherein, The word vector sequence determination submodel is used to: perform word segmentation processing on the input target item information and the target search keyword respectively, and determine a first word vector sequence corresponding to the target item information and a second word vector sequence corresponding to the target search keyword, and input the first word vector sequence and the second word vector sequence into the preset recurrent network submodel; The preset recurrent network sub-model is used to: perform vector integration on the input first word vector sequence and the second word vector sequence, determine the first text vector corresponding to the target item information and the second text vector corresponding to the target search keyword, and input the first text vector and the second text vector into the matching degree determination sub-model; The matching degree determination sub-model is used to: perform vector matching on the input first text vector and the second text vector, determine the target matching degree between the target item information and the target search keyword, and output the target matching degree.
6. The method according to claim 5, characterized in that The preset semantic matching model also includes: a self-attention mechanism sub-model; Accordingly, the word vector sequence determination sub-model is further used to: input the first word vector sequence and the second word vector sequence into the self-attention mechanism sub-model; The self-attention mechanism sub-model is used to: perform attention weight calculation on the input first word vector sequence and the second word vector sequence, determine the first attention weight corresponding to each first word vector in the first word vector sequence, and the second attention weight corresponding to each second word vector in the second word vector sequence, and input each of the first attention weights and each of the second attention weights into the preset recurrent network sub-model; The preset recurrent network sub-model is specifically used to: based on the input first attention weights and the input second attention weights, perform vector integration on the input first word vector sequence and the second word vector sequence, determine the first text vector corresponding to the target item information and the second text vector corresponding to the target search keyword, and input the first text vector and the second text vector into the matching degree determination sub-model.
7. The method according to claim 1, characterized in that Based on the search exposure click rate of the item, each item in the item database is sampled to obtain multiple sample data, including: Determine the search exposure click-through rate for each item in the item database; Determine the item data whose search exposure click rate is greater than a first preset threshold as positive sample data, and determine the standard matching degree corresponding to the positive sample data to be 1; The item data whose search exposure click rate is less than a second preset threshold is determined as negative sample data, and the standard matching degree corresponding to the negative sample data is determined to be 0.
8. The method according to any one of claims 1 to 7, characterized in that: The target item information includes at least one of target item title information, target item attribute information and target item category information.
9. A label verification device, characterized in that: include: A target object label acquisition module is used to acquire target object information and target object labels to be verified; A target search keyword determination module, used to determine a target search keyword that matches the target item tag; An information input module, used to input the target item information and the target search keyword into a preset semantic matching model, wherein the preset semantic matching model is pre-trained based on sample item information, sample search keywords and a standard matching degree, and the standard matching degree is determined based on the search exposure click rate of the sample item; A verification result determination module is used to determine the target matching degree between the target item information and the target search keyword according to the output of the preset semantic matching model, and determine the verification result corresponding to the target item label according to the target matching degree, wherein the verification result refers to the result of whether the target item label matches the target item information; A preset semantic matching model training module is used to sample each item in the item database based on the search exposure click rate of the item to obtain multiple sample data, wherein the sample data includes sample item information, sample search keywords and standard matching degree; each of the sample data is input into the preset semantic matching model, and the output matching degree corresponding to each of the sample data is determined according to the output of the preset semantic matching model; a training error is determined according to the output matching degree and the corresponding standard matching degree; and the preset semantic matching model is trained by taking the training error reaching a convergence condition as the training target.
10. A device, characterized in that: The device comprises: one or more processors; A memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the label verification method as described in any one of claims 1-8.
11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the label verification method as described in any one of claims 1 to 8 is implemented.
