Method for determining subject words, apparatus for determining subject words, and readable storage medium

Abstract

The application provides a subject word determination method, a subject word determination device and a readable storage medium. The subject word determination method comprises the following steps: obtaining historical search information, determining a subject word set and a commodity title set according to the historical search information; combining the subject word set and the commodity title set to obtain a sample set; training a preset model through the sample set to obtain a target prediction model and a target subject word set; receiving a to-be-predicted commodity title, inputting the to-be-predicted commodity title into the target prediction model to obtain a first subject word; and outputting a subject word matched with the first subject word in the target subject word set. The preset model is trained through the sample set, the accuracy of the obtained target prediction model and target subject word set is improved, and the accuracy of the determined subject word is further improved. Meanwhile, automatic training of the preset model is realized, and the training efficiency of the preset model is improved.

Description

Technical Field

[0001] This invention relates to the field of subject heading determination technology, and more specifically, to a method for determining subject headings, a device for determining subject headings, and a readable storage medium. Background Technology

[0002] The growth of the e-commerce market has driven the development of related research, making it increasingly important to accurately identify product keywords, especially to improve the matching degree between identified keywords and products.

[0003] In the process of developing this invention, the inventors discovered that when determining product keywords using machine learning, the training of a pre-set model typically involves manual annotation. This method suffers from low training efficiency, and the subjective differences in perception among individuals lead to low accuracy in the training results. Existing methods cannot identify non-adjacent words in the product name as product keywords, nor can they identify words not appearing in the product name, resulting in low accuracy in determining the keywords. No effective technical means to solve these problems has yet been proposed in this field. Summary of the Invention

[0004] The present invention aims to solve at least one of the above-mentioned technical problems.

[0005] Therefore, the first objective of this invention is to provide a method for determining subject terms.

[0006] A second objective of the present invention is to provide a device for determining subject terms.

[0007] A third objective of this invention is to provide a readable storage medium.

[0008] To achieve the first objective of this invention, the technical solution of this invention provides a method for determining keyword terms, comprising: acquiring historical search information; determining a keyword set and a product title set based on the historical search information; combining the keyword set and the product title set to obtain a sample set; training a preset model using the sample set to obtain a target prediction model and a target keyword set; receiving a product title to be predicted; inputting the product title to be predicted into the target prediction model to obtain a first keyword; and outputting the keyword in the target keyword set that matches the first keyword.

[0009] This technical solution acquires historical search information and obtains a set of keyword terms and a set of product titles based on this information, thus processing the historical search information. The keyword term set and the product title set are combined, and the resulting sample set is used to train a pre-defined model, improving the accuracy of the trained target prediction model and the target keyword set, thereby increasing the accuracy of the determined keyword terms. Furthermore, this application can automatically train the pre-defined model using the sample set, saving training time and improving training efficiency. The product titles to be predicted are input into the target prediction model to obtain the first keyword, and the target keyword set is output, showing the keyword terms that match the first keyword, thus improving the reliability of the keyword determination method.

[0010] In addition, the technical solution provided by the above-described technical solution of the present invention may also have the following additional technical features:

[0011] In the above technical solution, historical search information includes keywords and product titles. Each product title corresponds to n keywords, where n is a positive integer greater than or equal to 1. Obtaining historical search information and determining the keyword set and product title set based on it specifically includes: obtaining keywords from the historical search information and forming a keyword set; obtaining product titles from the historical search information and forming a product title set; combining the keyword set and product title set to obtain a sample set, specifically including: obtaining product titles from the product title set and obtaining n keywords corresponding to the product titles from the keyword set; and combining the product titles and the n keywords corresponding to the product titles to obtain the sample set.

[0012] In this technical solution, the sample set consists of a set of product titles and sets of keywords corresponding to those product titles. The format of the sample set is (keywords, product titles). Training the preset model using the sample set—that is, training the samples using the sets of keywords and product titles—improves the accuracy of the resulting target prediction model and target keyword set, thereby enhancing the accuracy of the determined keywords. Furthermore, this embodiment achieves automated training of the preset model by using the sample set, reducing manual intervention, improving training efficiency, and enhancing the reliability of the keyword determination method.

[0013] In any of the above technical solutions, the product title and the n keywords corresponding to the product title are combined to obtain a sample set. Specifically, this includes: determining whether the keywords meet the data cleaning conditions; taking the keywords that meet the data cleaning conditions as valid keywords; and combining the product title and the valid keywords corresponding to the product title to obtain a sample set.

[0014] In this technical solution, after obtaining n keywords corresponding to the product title from the keyword set, it is determined whether the keywords meet the data cleaning conditions. Keywords that meet the data cleaning conditions are considered valid keywords. By combining the product title and the valid keywords corresponding to the product title separately, the interference caused by keywords to the sample set is reduced, the accuracy of the obtained sample set is improved, and thus the accuracy of the keyword determination method is improved.

[0015] In any of the above technical solutions, the target prediction model is obtained by training a preset model with a sample set, specifically including: determining a positive sample set and a negative sample set based on the sample set; and training the preset model with the positive sample set and the negative sample set to obtain the target prediction model and the target keyword set.

[0016] This technical solution determines the positive and negative sample sets through a sample set, and trains the preset model using the positive and negative sample sets. This improves the accuracy of the obtained target prediction model and target keyword set, thereby improving the reliability of the keyword determination method. At the same time, the logic is simple and easy to implement, simplifying the training process of the preset model and improving training efficiency.

[0017] In any of the above technical solutions, determining the positive sample set based on the sample set specifically includes: obtaining the occurrence frequency of n keywords corresponding to the same product title in the sample set; determining the keywords as positive sample keywords based on the occurrence frequency of the keywords corresponding to the same product title meeting a preset frequency condition; combining the product title and the positive sample keywords corresponding to the product title to obtain the positive sample set; wherein the number of positive sample keywords corresponding to any product title is at least one.

[0018] This technical solution combines product titles and corresponding positive sample keywords to obtain a positive sample set, improving the matching degree between keywords in the positive sample set and product titles, and ensuring the accuracy of the positive sample set. Furthermore, the positive sample set is used to train a preset model, improving the accuracy of the obtained target prediction model and target keyword set. Moreover, in this technical solution, each product title corresponds to at least one positive sample keyword, further ensuring the reliability of the obtained positive sample set.

[0019] In any of the above technical solutions, determining the negative sample set based on the sample set specifically includes: obtaining m keywords from the keyword set as negative sample keywords; combining any product title from the product title set with the m negative sample keywords to obtain the negative sample set; where m is a positive integer greater than or equal to 1.

[0020] In this technical solution, when determining the negative sample set using the sample set, m keywords are selected from the keyword set as negative sample keywords. Each product title in the product title set is then combined with one of the m negative sample keywords to obtain the negative sample set. This improves the accuracy of the obtained negative sample set. Furthermore, the negative sample set is used to train a pre-defined model, thereby improving the accuracy of the obtained target prediction model and the target keyword set. Moreover, in this technical solution, m is a positive integer greater than or equal to 1, further enhancing the reliability of the obtained negative sample set.

[0021] In any of the above technical solutions, any positive sample in the positive sample set carries first label information, and any negative sample in the negative sample set carries second label information. The preset model is trained using the positive sample set and the negative sample set to obtain the target prediction model and the target topic word set. Specifically, this includes: performing feature representation on any topic word and any product title in the positive sample set and the negative sample set; inputting the feature-represented topic words and feature-represented product titles into the preset neural network model to obtain vector representations of topic words and vector representations of product titles, wherein the set of vector representations of topic words is used as the target topic word set; and obtaining the target prediction model based on the similarity comparison results of the vector representations of topic words and vector representations of product titles, the first label information carried by any positive sample, and the second label information carried by any negative sample.

[0022] This technical solution performs feature representation on any keyword and any product title in the positive and negative sample sets, enabling the feature-represented keyword and product title to be processed by a pre-defined neural network model. The feature-represented keyword and product title are then input into the pre-defined neural network model, which processes them to improve the accuracy and applicability of the resulting vector representations of keyword and product title.

[0023] This technical solution uses the set of topic words represented by vectors output by the neural network model as the target topic word set, improving the accuracy and reliability of the obtained target topic word set. Based on the similarity comparison results between the vector-represented topic words and the vector-represented product titles, the first label information carried by any positive sample, and the second label information carried by any negative sample, a target prediction model is obtained, improving the accuracy of the obtained target prediction model and thus improving the accuracy of the obtained topic words. At the same time, the logic is simple and easy to implement, improving the efficiency of topic word determination and reducing the cost of topic word determination.

[0024] Furthermore, this technical solution processes the feature-represented topic words and feature-represented product titles through a preset neural network model, enabling the topic word determination method to use non-adjacent or absent words in the product title to be predicted as topic words, thereby further improving the reliability of the topic word determination method.

[0025] In any of the above technical solutions, the topic words in the output topic word set that match the first topic word specifically include: outputting the topic word in the target topic word set that has the highest similarity value to the first topic word.

[0026] In this technical solution, after obtaining the first topic word through the target prediction model, the topic word with the highest similarity value to the first topic word in the target topic word set is output, which improves the matching degree between the obtained topic word and the product title and further ensures the reliability of the topic word determination method.

[0027] To achieve the second objective of this invention, the technical solution of this invention provides a subject word determination apparatus, comprising: an acquisition device for acquiring historical search information and determining a subject word set and a product title set based on the historical search information; a combination device for combining the subject word set and the product title set to obtain a sample set; a training device for training a preset model using the sample set to obtain a target prediction model and a target subject word set; a receiving device for receiving product titles to be predicted; a processing device for inputting the product titles to be predicted into the target prediction model to obtain a first subject word; an output device for outputting subject words in the target subject word set that match the first subject word; and a control device for controlling the steps of the subject word determination method for implementing any of the above technical solutions.

[0028] The keyword determination device in this technical solution can implement the steps of the keyword determination method of any of the above technical solutions, and therefore has all the beneficial effects of any of the above technical solutions, which will not be elaborated here.

[0029] To achieve the third objective of this invention, the technical solution of this invention provides a readable storage medium storing a program. When the program is executed, it can implement the steps of the subject term determination method as described in any of the above technical solutions, and therefore has all the beneficial effects of any of the above technical solutions, which will not be elaborated here.

[0030] Additional aspects and advantages of the invention will become apparent in the following description or may be learned by practice of the invention. Attached Figure Description

[0031] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:

[0032] Figure 1 This is one of the flowcharts illustrating the steps of a method for determining subject terms in some embodiments of the present invention;

[0033] Figure 2 This is a second flowchart of the method for determining the subject terms in some embodiments of the present invention;

[0034] Figure 3 This is the third flowchart of the method for determining the subject terms in some embodiments of the present invention;

[0035] Figure 4 This is the fourth flowchart of the method for determining the subject terms in some embodiments of the present invention;

[0036] Figure 5 This is the fifth flowchart of the method for determining the subject terms in some embodiments of the present invention;

[0037] Figure 6 This is the sixth flowchart of the method for determining the subject terms in some embodiments of the present invention;

[0038] Figure 7 This is the seventh flowchart of the method for determining the subject terms in some embodiments of the present invention;

[0039] Figure 8 This is the eighth flowchart of the method for determining the subject terms in some embodiments of the present invention;

[0040] Figure 9 This is a schematic block diagram of the structure of a subject term determination device according to some embodiments of the present invention;

[0041] Figure 10 This is the ninth flowchart of the method for determining the subject terms in some embodiments of the present invention;

[0042] Figure 11 This is one of the schematic diagrams illustrating the principle of the method for determining subject terms in some embodiments of the present invention.

[0043] Figure 12 This is the second schematic diagram illustrating the principle of the method for determining the subject terms in some embodiments of the present invention.

[0044] in, Figure 9 The correspondence between the reference numerals and component names in the attached drawings is as follows:

[0045] 100: Keyword determination device; 110: Acquisition device; 120: Combination device; 130: Training device; 140: Receiving device; 150: Processing device; 160: Output device; 170: Control device. Detailed Implementation

[0046] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0047] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0048] The following reference Figures 1 to 12 This invention describes a method for determining subject terms, a subject term determination apparatus 100, and a readable storage medium according to some embodiments of the present invention.

[0049] Example 1:

[0050] like Figure 1 As shown, this embodiment provides a method for determining keywords, including:

[0051] Step S102: Obtain historical search information, and determine the set of keywords and the set of product titles based on the historical search information;

[0052] Step S104: Combine the keyword set and the product title set to obtain the sample set;

[0053] Step S106: Train the preset model using the sample set to obtain the target prediction model and the target topic word set;

[0054] Step S108: Receive the product title to be predicted, input the product title to be predicted into the target prediction model, and obtain the first topic keyword;

[0055] Step S110: Output the keywords in the target keyword set that match the first keyword.

[0056] Understandably, in this embodiment, the keyword can be a word or phrase used to represent the theme of the product name. The keyword set and product title set are determined through historical search information. This historical search information can be the user's search information over a certain period, including search terms, search results, clicked product titles, and product reviews, etc. In this embodiment, the historical search information can be obtained through the user's search log.

[0057] In some implementations of this embodiment, historical search information may be the search information of a certain user within a specific time period, the search information of a certain user group within a specific time period, or the search information of all users within a specific time period.

[0058] This embodiment acquires historical search information and obtains a set of keyword terms and a set of product titles based on this information, thus processing the historical search information. The keyword term set and the product title set are combined, and the resulting sample set is used to train a preset model, improving the accuracy of the trained target prediction model and the target keyword set, thereby improving the accuracy of the determined keyword terms. Furthermore, this application can automatically train the preset model using the sample set, saving training time and improving training efficiency. The product titles to be predicted are input into the target prediction model to obtain the first keyword, and the keyword terms matching the first keyword in the target keyword set are output, improving the reliability of the keyword determination method.

[0059] In some implementations of this embodiment, the keyword set can be a set of all keywords in the historical search information, and the product title set can be a set of all product titles in the historical search information.

[0060] In other embodiments of this example, the number of keywords matching the first keyword can be one or more.

[0061] Example 2:

[0062] like Figure 2 As shown, this embodiment provides a method for determining keywords. In addition to the technical features of Embodiment 1 above, this embodiment further includes the following technical features.

[0063] Historical search information includes keywords and product titles. Each product title corresponds to n keywords, where n is a positive integer greater than or equal to 1. The process involves obtaining historical search information and determining the keyword set and product title set based on it. Specifically, this includes: obtaining keywords from the historical search information and forming a keyword set; obtaining product titles from the historical search information and forming a product title set; and combining the keyword set and product title set to obtain a sample set. This involves: obtaining product titles from the product title set and obtaining the n keywords corresponding to each product title from the keyword set; and combining the product titles and their corresponding n keywords to obtain the sample set.

[0064] In other words, the method for determining keywords in this embodiment specifically includes:

[0065] Step S202: Obtain the keywords from historical search information and form a keyword set using the keywords;

[0066] Step S204: Obtain product titles from historical search information and assemble a product title set using these product titles;

[0067] Step S206: Obtain the product title from the product title set and obtain n keywords corresponding to the product title from the keyword set;

[0068] Step S208: Combine the product title and the n keywords corresponding to the product title to obtain the sample set;

[0069] Step S210: Train the preset model using the sample set to obtain the target prediction model and the target topic word set;

[0070] Step S212: Receive the product title to be predicted, input the product title to be predicted into the target prediction model, and obtain the first topic keyword;

[0071] Step S214: Output the keywords in the target keyword set that match the first keyword.

[0072] In this embodiment, the historical search information includes keywords and product titles. It can be understood that n keywords can be determined through any product title. Therefore, any product title can correspond to n keywords, where n is a positive integer greater than or equal to 1.

[0073] In this embodiment, keywords from historical search information are obtained, and a keyword set is formed using these keywords. Product titles from historical search information are also obtained, and a product title set is formed using these product titles. Product titles are then retrieved from the product title set, and n keywords corresponding to the product titles are retrieved from the keyword set. The product titles and their corresponding n keywords are combined to obtain a sample set.

[0074] In this embodiment, the sample set consists of a set of product titles and a set of keywords corresponding to those product titles. The format of the sample set is (keywords, product titles). Training the preset model using this sample set—that is, training the sample set with the keywords and product titles—improves the accuracy of the resulting target prediction model and target keyword set, thereby enhancing the accuracy of the determined keywords. Furthermore, this embodiment achieves automated training of the preset model using the sample set, reducing manual intervention, improving training efficiency, and enhancing the reliability of the keyword determination method.

[0075] In some implementations of this embodiment, n can take the values ​​of 3, 5, or 10.

[0076] In other implementations of this embodiment, the number of times each keyword appears in the historical search information can be one or more.

[0077] Example 3:

[0078] like Figure 3 As shown, this embodiment provides a method for determining keywords. In addition to the technical features of any of the above embodiments, this embodiment further includes the following technical features.

[0079] The product title and the n corresponding keywords are combined to obtain a sample set. Specifically, this includes: determining whether the keywords meet the data cleaning conditions; taking the keywords that meet the data cleaning conditions as valid keywords; and combining the product title and the valid keywords corresponding to the product title to obtain a sample set.

[0080] In other words, the method for determining keywords in this embodiment specifically includes:

[0081] Step S302: Obtain the keywords from historical search information and form a keyword set using the keywords;

[0082] Step S304: Obtain product titles from historical search information and assemble a product title set using these product titles;

[0083] Step S306: Obtain the product title from the product title set and obtain n keywords corresponding to the product title from the keyword set;

[0084] Step S308: Determine whether the keywords meet the data cleaning conditions; keywords that meet the data cleaning conditions are considered valid keywords.

[0085] Step S310: Combine the product title and the valid keywords corresponding to the product title to obtain a sample set;

[0086] Step S312: Train the preset model using the sample set to obtain the target prediction model and the target topic word set;

[0087] Step S314: Receive the product title to be predicted, input the product title to be predicted into the target prediction model, and obtain the first topic keyword;

[0088] Step S316: Output the keywords in the target keyword set that match the first keyword.

[0089] In this embodiment, after obtaining n keywords corresponding to the product title from the keyword set, it is determined whether the keywords meet the data cleaning conditions. Keywords that meet the data cleaning conditions are considered valid keywords. By combining the product title and the valid keywords corresponding to the product title separately, the interference caused by keywords to the sample set is reduced, the accuracy of the obtained sample set is improved, and thus the reliability of the keyword determination method is improved.

[0090] In some embodiments of this embodiment, the data cleaning condition may be the occurrence frequency of the subject terms, and the subject terms with a higher occurrence frequency are used as valid subject terms. The data cleaning condition may also be the length of the subject terms, and the subject terms with a length within the preset length range are used as valid subject terms. The preset length range may be from 2 to 7. The data cleaning condition may also be to delete stop words, such as deleting words without practical meaning like "的", "了", "the", "is", "at", or "on".

[0091] Embodiment 4:

[0092] As Figure 4 shown, this embodiment provides a method for determining subject terms. In addition to the technical features of any of the above embodiments, this embodiment further includes the following technical features.

[0093] Training a preset model with a sample set to obtain a target prediction model, specifically including: determining a positive sample set and a negative sample set according to the sample set; training the preset model with the positive sample set and the negative sample set to obtain a target prediction model and a target subject term set.

[0094] In other words, the method for determining subject terms in this embodiment specifically includes:

[0095] Step S402, obtaining historical search information, and determining a subject term set and a product title set according to the historical search information;

[0096] Step S404, combining the subject term set and the product title set to obtain a sample set;

[0097] Step S406, determining a positive sample set and a negative sample set according to the sample set;

[0098] Step S408, training the preset model with the positive sample set and the negative sample set to obtain a target prediction model and a target subject term set;

[0099] Step S410, receiving a product title to be predicted, inputting the product title to be predicted into the target prediction model to obtain a first subject term;

[0100] Step S412, outputting the subject terms in the target subject term set that match the first subject term.

[0101] In this embodiment, a positive sample set and a negative sample set are determined according to the sample set. It can be understood that in the positive sample set, the matching degree between the product title and the subject term is relatively high, while in the negative sample set, the matching degree between the product title and the subject term is relatively low.

[0102] This embodiment determines the positive and negative sample sets through the sample set, and trains the preset model using the positive and negative sample sets. This improves the accuracy of the obtained target prediction model and target keyword set, thereby improving the reliability of the keyword determination method. At the same time, the logic is simple and easy to implement, simplifying the training process of the preset model and improving training efficiency.

[0103] In some embodiments of this example, the number of negative samples can be greater than the number of positive samples, further ensuring the accuracy of the obtained target prediction model and target keyword set. In other embodiments of this example, the number of negative samples can be five times the number of positive samples.

[0104] Example 5:

[0105] like Figure 5 As shown, this embodiment provides a method for determining keywords. In addition to the technical features of any of the above embodiments, this embodiment further includes the following technical features.

[0106] Based on the sample set, the positive sample set is determined, specifically including: obtaining the occurrence frequency of n keywords corresponding to the same product title in the sample set; determining the keywords as positive sample keywords based on the occurrence frequency of the keywords corresponding to the same product title meeting the preset frequency conditions; combining the product title and the positive sample keywords corresponding to the product title to obtain the positive sample set; wherein, the number of positive sample keywords corresponding to any product title is at least one.

[0107] In other words, the method for determining keywords in this embodiment specifically includes:

[0108] Step S502: Obtain historical search information, and determine the set of keywords and the set of product titles based on the historical search information;

[0109] Step S504: Combine the keyword set and the product title set to obtain the sample set;

[0110] Step S506: Obtain the frequency of occurrence of n keywords corresponding to the same product title in the sample set;

[0111] Step S508: Based on the fact that the frequency of occurrence of the keywords corresponding to the same product title meets the preset frequency conditions, the keywords are determined to be positive sample keywords;

[0112] Step S510: Combine the product title and the positive sample keywords corresponding to the product title to obtain the positive sample set;

[0113] Step S512: Determine the negative sample set based on the sample set;

[0114] Step S514: Train the preset model using the positive sample set and the negative sample set to obtain the target prediction model and the target topic word set;

[0115] Step S516: Receive the product title to be predicted, input the product title to be predicted into the target prediction model, and obtain the first topic keyword;

[0116] Step S518: Output the keywords in the target keyword set that match the first keyword.

[0117] In this embodiment, when determining the positive sample set through the sample set, the frequency of occurrence of n keywords corresponding to the same product title in the sample set is obtained. Understandably, the frequency of occurrence of the n keywords corresponding to the same product title can be obtained by dividing the number of occurrences of each keyword corresponding to the same product title by the total number of occurrences of all keywords corresponding to the same product title. When the frequency of occurrence of keywords corresponding to the same product title meets a preset frequency condition, the keywords that meet the preset frequency condition are determined as positive sample keywords.

[0118] In some embodiments of this example, the preset frequency condition can be the relationship between the occurrence frequency of n keywords corresponding to the same product and a set threshold, and keywords with an occurrence frequency greater than the set threshold are used as positive sample keywords.

[0119] In other embodiments of this example, the frequency of occurrence of n keywords corresponding to the same product can be arranged from smallest to largest, with the preset frequency condition being that the top preset number of keywords are used as positive sample keywords. Understandably, the preset number can be 3 or 5.

[0120] This embodiment combines the product title and the corresponding positive sample keywords to obtain a positive sample set, improving the matching degree between the keywords in the positive sample set and the product title, and ensuring the accuracy of the positive sample set. Furthermore, the positive sample set is used to train a preset model, improving the accuracy of the obtained target prediction model and target keyword set. In addition, in this embodiment, the number of positive sample keywords corresponding to any product title is at least one, further ensuring the reliability of the obtained positive sample set.

[0121] Example 6:

[0122] like Figure 6 As shown, this embodiment provides a method for determining keywords. In addition to the technical features of any of the above embodiments, this embodiment further includes the following technical features.

[0123] Based on the sample set, the negative sample set is determined, specifically including: obtaining m keywords from the keyword set as negative sample keywords; combining any product title from the product title set with the m negative sample keywords to obtain the negative sample set; where m is a positive integer greater than or equal to 1.

[0124] In other words, the method for determining keywords in this embodiment specifically includes:

[0125] Step S602: Obtain historical search information, and determine the set of keywords and the set of product titles based on the historical search information;

[0126] Step S604: Combine the keyword set and the product title set to obtain the sample set;

[0127] Step S606: Obtain the frequency of occurrence of n keywords corresponding to the same product title in the sample set;

[0128] Step S608: Based on the fact that the frequency of occurrence of the keywords corresponding to the same product title meets the preset frequency conditions, the keywords are determined to be positive sample keywords;

[0129] Step S610: Combine the product title and the positive sample keywords corresponding to the product title to obtain a positive sample set;

[0130] Step S612: In the keyword set, obtain m keywords as negative sample keywords;

[0131] Step S614: Combine any product title in the product title set with m negative sample keywords to obtain a negative sample set;

[0132] Step S616: Train the preset model using the positive sample set and the negative sample set to obtain the target prediction model and the target topic word set;

[0133] Step S618: Receive the product title to be predicted, input the product title to be predicted into the target prediction model, and obtain the first topic keyword;

[0134] Step S620: Output the keywords in the target keyword set that match the first keyword.

[0135] In this embodiment, when determining the negative sample set using the sample set, m keywords are selected from the keyword set as negative sample keywords. Any product title from the product title set is combined with the m negative sample keywords to obtain the negative sample set, thus improving the accuracy of the obtained negative sample set. Furthermore, the negative sample set is used to train the preset model, improving the accuracy of the obtained target prediction model and target keyword set. In this embodiment, m is a positive integer greater than or equal to 1, further enhancing the reliability of the obtained negative sample set.

[0136] In some implementations of this embodiment, m keywords can be randomly selected from the keyword set as negative sample keywords to improve the efficiency of determining negative sample keywords, thereby improving the efficiency of the keyword determination method.

[0137] Example 7:

[0138] like Figure 7 As shown, this embodiment provides a method for determining keywords. In addition to the technical features of any of the above embodiments, this embodiment further includes the following technical features.

[0139] Each positive sample in the positive sample set carries a first label information, and each negative sample in the negative sample set carries a second label information. The preset model is trained using the positive and negative sample sets to obtain the target prediction model and the target topic word set. Specifically, this includes: performing feature representation on any topic word and any product title in the positive and negative sample sets; inputting the feature-represented topic words and feature-represented product titles into the preset neural network model to obtain vector representations of topic words and vector representations of product titles, wherein the set of vector representations of topic words is used as the target topic word set; and obtaining the target prediction model based on the similarity comparison results of the vector representations of topic words and vector representations of product titles, the first label information carried by any positive sample, and the second label information carried by any negative sample.

[0140] In other words, the method for determining keywords in this embodiment specifically includes:

[0141] Step S702: Obtain historical search information, and determine the set of keywords and the set of product titles based on the historical search information;

[0142] Step S704: Combine the keyword set and the product title set to obtain the sample set;

[0143] Step S706: Determine the positive sample set and the negative sample set based on the sample set;

[0144] Step S708: Perform feature representation on any keyword and any product title in the positive sample set and the negative sample set;

[0145] Step S710: Input the feature-represented topic words and feature-represented product titles into a preset neural network model to obtain vector-represented topic words and vector-represented product titles, wherein the set of vector-represented topic words is used as the target topic word set;

[0146] Step S712: Based on the similarity comparison results between the topic words represented by vectors and the product titles represented by vectors, the first label information carried by any positive sample and the second label information carried by any negative sample, the target prediction model is obtained.

[0147] Step S714: Receive the title of the product to be predicted, input the title of the product to be predicted into the target prediction model, and obtain the first topic word;

[0148] Step S716: Output the keywords in the target keyword set that match the first keyword.

[0149] In this embodiment, each positive sample in the positive sample set carries a first label information, and each negative sample in the negative sample set carries a second label information. Understandably, the first label information is used to represent the degree of matching between the product title and the keyword in the positive sample set. The second label information is used to identify the degree of matching between the product title and the keyword in the negative sample set.

[0150] In some implementations of this embodiment, the first tag information and the second tag information can be represented by 1 and 0, where 1 indicates a match between the product title and the keyword, and 0 indicates a mismatch between the product title and the keyword.

[0151] This embodiment performs feature representation on any keyword and any product title in the positive and negative sample sets, enabling the feature-represented keyword and product title to be processed by a preset neural network model. Understandably, this embodiment can use one-hot encoding or mathematical vectors to feature-identify any keyword and any product title.

[0152] The feature-represented keywords and product titles are input into a preset neural network model. This model processes the feature-labeled keywords and product titles, improving the accuracy and applicability of the resulting vector representations of keywords and product titles. Understandably, the preset neural network model in this embodiment can be a multi-layer neural network model, further ensuring the reliability of the obtained vector-labeled keywords and product titles.

[0153] This embodiment uses the set of topic words represented by vectors output by the neural network model as the target topic word set, which improves the accuracy and reliability of the obtained target topic word set.

[0154] This embodiment obtains a target prediction model based on the similarity comparison results between the vector-represented keywords and the vector-represented product titles, the first label information carried by any positive sample, and the second label information carried by any negative sample. This improves the accuracy of the obtained target prediction model, thereby improving the accuracy of the obtained keywords. At the same time, the logic is simple and easy to implement, improving the efficiency of keyword determination and reducing the cost of keyword determination.

[0155] Furthermore, this embodiment trains the preset model using positive and negative sample sets, and processes the feature-represented topic words and feature-represented product titles using a preset neural network model. This enables the topic word determination method to use non-adjacent or non-occurring words in the product title to be predicted as topic words, further improving the reliability of the topic word determination method.

[0156] In some implementations of this embodiment, the similarity comparison results between the vector representation of topic words and the vector representation of product titles can be obtained through distance functions, such as Euclidean distance function or cosine distance function.

[0157] Example 8:

[0158] like Figure 8 As shown, this embodiment provides a method for determining keywords. In addition to the technical features of any of the above embodiments, this embodiment further includes the following technical features.

[0159] Output the keywords in the target keyword set that match the first keyword, specifically including: output the keyword in the target keyword set that has the highest similarity score to the first keyword.

[0160] In other words, the method for determining keywords in this embodiment specifically includes:

[0161] Step S802: Obtain historical search information, and determine the set of keywords and the set of product titles based on the historical search information;

[0162] Step S804: Combine the keyword set and the product title set to obtain the sample set;

[0163] Step S806: Train the preset model using the sample set to obtain the target prediction model and the target topic word set;

[0164] Step S808: Receive the product title to be predicted, input the product title to be predicted into the target prediction model, and obtain the first topic keyword;

[0165] Step S810: Output the keyword in the target keyword set that has the highest similarity score to the first keyword.

[0166] In this embodiment, after obtaining the first topic word through the target prediction model, the topic word with the highest similarity value to the first topic word in the target topic word set is output, which improves the matching degree between the obtained topic word and the product title and further ensures the reliability of the topic word determination method.

[0167] Example 9:

[0168] like Figure 9 As shown, this embodiment provides a topic word determination device 100, including: an acquisition device 110, a combination device 120, a training device 130, a receiving device 140, a processing device 150, an output device 160, and a control device 170. The acquisition device 110 is used to acquire historical search information and determine a topic word set and a product title set based on the historical search information. The combination device 120 combines the topic word set and the product title set to obtain a sample set. The training device 130 is used to train a preset model using the sample set to obtain a target prediction model and a target topic word set. The receiving device 140 is used to receive the product title to be predicted. The processing device 150 is used to input the product title to be predicted into the target prediction model to obtain a first topic word. The output device 160 is used to output the topic words in the target topic word set that match the first topic word. The control device 170 controls the steps for implementing the topic word determination method of any of the above embodiments, thus possessing all the beneficial effects of any of the above embodiments, which will not be elaborated further here.

[0169] Example 10:

[0170] This embodiment provides a readable storage medium storing a program. When the program is executed, it can implement the steps of the subject term determination method as described in any of the above embodiments. Therefore, it has all the beneficial effects of any of the above embodiments, which will not be repeated here.

[0171] Example 11:

[0172] In related technologies, when determining the keywords of a product through machine learning, the preset model is usually trained by manual annotation, which reduces the training efficiency of the preset model. Moreover, the differences in subjective cognition among different people reduce the accuracy of the training results.

[0173] Furthermore, in determining the keywords of a product, the related technologies cannot use non-adjacent words in the product name as keywords, nor can they use words that do not appear in the product name as keywords, thus reducing the accuracy of the determined keywords.

[0174] For example, the product name is "OLT (OLT) Bed Technology Fabric Bed No-Wash Fabric Bed Nordic Modern Bedroom Furniture Double 1.5m 1.8m Soft Bed Nano Waterproof Technology Fabric [Single Bed] Frame Bed"

[0175] For a product with dimensions of 1500*2000 mm, the corresponding keywords should be "double bed," "floor-to-ceiling bed," or "fabric bed," etc. However, since the word "double" and "bed" are not adjacent in the product name, and the word "floor-to-ceiling bed" does not appear in the product name, the determination method in the related technology cannot use "double bed" or "floor-to-ceiling bed" as the product's keywords, thus reducing the reliability of the keyword determination method.

[0176] To address the aforementioned problems in related technologies, such as Figure 10 As shown, this embodiment provides a method for determining keywords, specifically including:

[0177] Step S902: Obtain the keywords from historical search information and form a keyword set using the keywords;

[0178] Step S904: Obtain product titles from historical search information and assemble a product title set using these product titles;

[0179] Step S906: Obtain the product title from the product title set and obtain n keywords corresponding to the product title from the keyword set;

[0180] Step S908: Determine whether the keywords meet the data cleaning conditions; keywords that meet the data cleaning conditions are considered valid keywords.

[0181] Step S910: Combine the product title and the valid keywords corresponding to the product title to obtain a sample set;

[0182] Step S912: Obtain the frequency of occurrence of n keywords corresponding to the same product title in the sample set;

[0183] Step S914: Based on the fact that the frequency of occurrence of the keywords corresponding to the same product title meets the preset frequency conditions, the keywords are determined to be positive sample keywords.

[0184] Step S916: Combine the product title and the positive sample keywords corresponding to the product title to obtain the positive sample set;

[0185] Step S918: In the keyword set, obtain m keywords as negative sample keywords;

[0186] Step S920: Combine any product title in the product title set with m negative sample keywords to obtain a negative sample set;

[0187] Step S922: Perform feature representation on any keyword and any product title in the positive sample set and the negative sample set;

[0188] Step S924: Input the feature-represented topic words and feature-represented product titles into a preset neural network model to obtain vector-represented topic words and vector-represented product titles, wherein the set of vector-represented topic words is used as the target topic word set;

[0189] Step S926: Based on the similarity comparison results between the topic words represented by vectors and the product titles represented by vectors, the first label information carried by any positive sample and the second label information carried by any negative sample, the target prediction model is obtained.

[0190] Step S928: Output the keyword in the target keyword set that has the highest similarity score to the first keyword.

[0191] like Figure 11 and Figure 12 As shown, this embodiment obtains a set of keywords and a set of product titles based on historical search information. The keyword set is then cleaned using data cleaning conditions to obtain valid keywords. The product titles and their corresponding valid keywords are then combined to obtain a sample set. Understandably, the format of the sample set in this embodiment is (keywords, product titles).

[0192] In some implementations of this embodiment, data cleaning conditions can include deleting stop words, deleting words longer than 7, deleting words shorter than 2, or deleting low-frequency words, thereby improving the accuracy of the obtained effective keywords.

[0193] This embodiment determines a positive sample set and a negative sample set based on the sample set. In the positive sample set, the keywords match the product titles more closely, while in the negative sample set, the keywords match the product titles less closely.

[0194] Specifically, the frequency of occurrence of n keywords corresponding to the same product title in the sample set is obtained, and the keywords whose frequency meets the preset frequency condition are determined as positive sample keywords. The product title and the positive sample keywords corresponding to the product title are combined to obtain the positive sample set. In some embodiments of this example, the three most frequently occurring keywords corresponding to any product title can be retained as positive sample keywords.

[0195] In this embodiment, m keywords are randomly selected from the keyword set as negative sample keywords. Any product title from the product title set is combined with the m negative sample keywords to obtain a negative sample set. In some implementations of this embodiment, five keywords can be randomly selected from the keyword set as negative sample keywords. In other implementations of this embodiment, the number of keywords in the negative sample set can be five times the number of keywords in the positive sample set.

[0196] This embodiment trains a preset model using positive and negative samples. Specifically, it performs feature representation on any topic word and any product title in the positive and negative sample sets. The feature-represented topic words and product titles are then input into the preset neural network model to obtain vector representations of topic words and product titles. The set of vector-represented topic words serves as the target topic word set.

[0197] In some embodiments of this example, the feature-represented keyword and the feature-represented product title can also be input into the Seq2Seq model (also known as the "sequence-to-sequence model"), which improves the flexibility of the keyword determination method.

[0198] This embodiment obtains the target prediction model based on the similarity comparison results of the vector-represented topic words and the vector-represented product titles, the first label information carried by any positive sample, and the second label information carried by any negative sample. The logic is simple and easy to implement, which improves the reliability of the topic word determination method.

[0199] In some embodiments of this example, the similarity comparison results between the vector-represented keywords and the vector-represented product titles can be obtained through a distance function, such as the Euclidean distance function or the cosine distance function.

[0200] In summary, the beneficial effects of the embodiments of the present invention are as follows:

[0201] 1. This invention obtains a positive sample set and a negative sample set by searching historical information, and trains a preset model using the positive sample set and the negative sample set, thereby improving the accuracy of the obtained target prediction model and the target topic word set, and thus improving the accuracy of the obtained topic words;

[0202] 2. This invention enables automatic training of a preset model based on a set of positive samples and a set of negative samples, which improves the training efficiency of the preset model and eliminates the need for manual intervention, thereby improving the reliability of the training of the preset model.

[0203] 3. This invention trains a preset model using positive and negative sample sets, and processes the feature-represented keywords and product titles using a neural network model. This enables the keyword determination method to use non-adjacent or absent words in the product title as keywords, thereby improving the reliability of the keyword determination method.

[0204] In this invention, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance; the term "multiple" refers to two or more unless otherwise explicitly defined. The terms "install," "connect," "link," and "fix" should be interpreted broadly. For example, "connect" can be a fixed connection, a detachable connection, or an integral connection; "link" can be a direct connection or an indirect connection through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0205] In the description of this invention, it should be understood that the terms "upper," "lower," "left," "right," "front," "rear," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or unit referred to must have a specific orientation or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0206] In the description of this specification, the terms "one embodiment," "some embodiments," "specific embodiment," etc., refer to a specific feature, structure, material, or characteristic described in connection with that embodiment or example, which is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0207] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for determining keyword terms, characterized in that, include: Obtain historical search information, and determine a set of keywords and a set of product titles based on the historical search information. The historical search information includes keywords and product titles, and any product title corresponds to n keywords, where n is a positive integer greater than or equal to 1. The keyword set and the product title set are combined to obtain a sample set. Based on the sample set, a positive sample set and a negative sample set are determined. Any positive sample in the positive sample set carries a first tag information, and any negative sample in the negative sample set carries a second tag information. The preset model is trained using the sample set to obtain a target prediction model and a target keyword set, specifically including: Perform feature representation on any keyword and any product title in the positive sample set and the negative sample set; The topic words and product titles represented by the features are input into a preset neural network model to obtain the topic words and product titles represented by vectors, wherein the set of topic words represented by vectors is used as the target topic word set; The target prediction model is obtained based on the similarity comparison results of the topic words represented by vectors and the product titles represented by vectors, the first tag information carried by any positive sample and the second tag information carried by any negative sample; Receive the product title to be predicted, input the product title to be predicted into the target prediction model, and obtain the first topic keyword; Output the keywords in the target keyword set that match the first keyword.

2. The method for determining keywords according to claim 1, characterized in that, The step of obtaining the historical search information and determining the keyword set and the product title set based on the historical search information specifically includes: Obtain the keywords from the historical search information, and form the keyword set using the keywords; Obtain the product titles from the historical search information, and assemble the product title set using the product titles; The process of combining the keyword set and the product title set to obtain the sample set specifically includes: Obtain the product title from the product title set, and obtain n keywords corresponding to the product title from the keyword set; The sample set is obtained by combining the product title and the n keywords corresponding to the product title.

3. The method for determining keywords according to claim 2, characterized in that, The step of combining the product title and the n keywords corresponding to the product title to obtain the sample set specifically includes: Determine whether the keywords meet the data cleaning conditions; The keywords that meet the data cleaning conditions are taken as valid keywords; The sample set is obtained by combining the product title and the corresponding effective keywords.

4. The method for determining keywords according to claim 2, characterized in that, The step of training the preset model using the sample set to obtain the target prediction model specifically includes: The preset model is trained using the positive sample set and the negative sample set to obtain the target prediction model and the target topic word set.

5. The method for determining keywords according to claim 4, characterized in that, The step of determining the positive sample set based on the sample set specifically includes: Obtain the frequency of occurrence of n keywords corresponding to the same product title in the sample set; Based on the fact that the frequency of occurrence of the keyword corresponding to the same product title meets a preset frequency condition, the keyword is determined to be a positive sample keyword; The product title and the corresponding positive sample keywords are combined to obtain the positive sample set; Among them, the number of positive sample keywords corresponding to any one of the product titles is at least one.

6. The method for determining keywords according to claim 4, characterized in that, The step of determining the negative sample set based on the sample set specifically includes: From the set of keywords, m keywords are selected as negative sample keywords; The negative sample set is obtained by combining any one of the product titles in the product title set with m negative sample keywords. Where m is a positive integer greater than or equal to 1.

7. The method for determining subject terms according to any one of claims 1 to 6, characterized in that, The step of outputting the topic terms in the topic term set that match the first topic term specifically includes: Output the keyword in the target keyword set that has the highest similarity score to the first keyword.

8. A subject term determination device (100), characterized in that, include: Acquisition device (110) is used to acquire historical search information and determine a set of keywords and a set of product titles based on the historical search information; The combining device (120) combines the keyword set and the product title set to obtain a sample set; Training device (130) is used to train a preset model using the sample set to obtain a target prediction model and a target keyword set; Receiving device (140) for receiving the title of the product to be predicted; Processing device (150) is used to input the title of the product to be predicted into the target prediction model to obtain the first topic word; Output device (160) is used to output the subject words in the target subject word set that match the first subject word; A control device (170) is used to implement the steps of the method for determining subject terms as described in any one of claims 1 to 7.

9. A readable storage medium, characterized in that, The storage medium stores a program that, when executed, enables the implementation of the steps of the subject term determination method as described in any one of claims 1 to 7.

Images