CN116579825A - Category prediction method, device, equipment and storage medium - Google Patents

Abstract

The invention discloses a category prediction method, a category prediction device, category prediction equipment and a storage medium; in the scheme, category correspondence needs to be predetermined, wherein the category correspondence comprises a first category correspondence between keywords and categories, a second category correspondence between vocabularies of commodities in a commodity library and categories, and a third category correspondence between vocabularies of commodities which are successfully transacted in a predetermined period, which are determined through historical click behavior data; when the category prediction is carried out, the category can be predicted according to the first category corresponding relation determined by the historical click behavior data, and the corresponding relation prediction of the vocabulary of the commodity in the commodity library and the vocabulary of the commodity successfully transacted in the preset period can be utilized, so that the category prediction can be accurately carried out no matter whether the category has click behavior or not, and further the accuracy of commodity recall is improved.

Description

Category prediction method, device, equipment and storage medium

technical field

The present invention relates to the technical field of category prediction, and more specifically, to a category prediction method, device, equipment and storage medium.

Background technique

At present, on the e-commerce platform, the supplier creates a product, and needs to check whether the product can be searched by the user on the e-commerce platform. Usually, the keyword entered by the user can be matched to the corresponding category. For example, if the user enters the keyword [computer chair], it can be matched to the category of [office], so the recalled products are all products related to computer chairs. In this case, the search quality is better.

However, in many cases, because the keywords entered by the user cannot match the corresponding category, the corresponding product cannot be searched, thereby reducing the user experience and making the transaction impossible to complete. For example: the user enters the keyword [computer], if the category of [digital] is not matched but the category of [office] is matched, then the products such as computer desks and computer accessories are recalled. This search solution has poor search quality and poor user experience, which reduces the probability of closing a transaction.

Therefore, how to accurately predict categories is a problem to be solved by those skilled in the art.

Contents of the invention

The object of the present invention is to provide a category prediction method, device, equipment and storage medium to improve the accuracy of category prediction.

In order to achieve the above object, the present invention provides a category prediction method, including:

Determine the target keyword entered by the user;

Matching the target keyword with the category correspondence to determine the predicted category;

Wherein, the category correspondence includes: the first category correspondence between keywords and categories determined through historical click behavior data, and/or the second category correspondence between commodity vocabulary and categories in the commodity library. A category correspondence, and/or a third category correspondence between vocabulary and categories of commodities that are successfully traded within a predetermined period.

Preferably, before determining the target keyword input by the user, it also includes:

Obtain the historical click behavior data of each keyword under the first category of each recalled product;

Determine the number of clicks of each keyword under each first category according to the historical click behavior data, and the total number of clicks of each keyword;

Using the number of clicks of each keyword under each first category and the total number of clicks of each keyword, the first weight value of each keyword is calculated to generate a first category correspondence.

Preferably, before determining the number of clicks of each keyword under each first category according to the historical click behavior data, the method further includes:

Delete invalid keywords in the historical click behavior data; the invalid keywords include: keywords with special characters and keywords with a length of 1.

Preferably, after the calculation of the first weight value of each keyword, it also includes:

Detecting whether the weight value of each keyword under each first category is less than a predetermined threshold;

If yes, delete the corresponding relationship between the keywords whose weight value is less than the predetermined threshold and the first category.

Preferably, before determining the target keyword input by the user, it also includes:

Obtain the first basic information of each commodity in the commodity library; the first basic information includes commodity title and attribute;

splitting each first basic information into a plurality of first commodity words;

Determining the co-occurrence times that each first commodity vocabulary and each second category in each first basic information co-occur in the first basic information of the same commodity;

determining the total number of occurrences of each first commodity term;

The second weight value of each first commodity vocabulary under each second category is calculated by using the co-occurrence times and total occurrence times of each first commodity vocabulary and each second category to generate a second category correspondence.

Preferably, before determining the target keyword input by the user, it also includes:

Acquiring second basic information of commodities successfully traded within a predetermined period; the second basic information includes commodity titles and attributes;

splitting each second basic information into a plurality of second commodity words;

Determining the co-occurrence times that each second commodity vocabulary and each third category in each second basic information co-occur in the first basic information of the same commodity;

determining the total number of occurrences of each second commodity term;

The third weight value of each second commodity vocabulary under each third category is calculated by using the co-occurrence times and total occurrence times of each second commodity vocabulary and each third category to generate a third category correspondence.

Preferably, said determining the target keyword input by the user includes:

Get the search term entered by the user;

A plurality of target keywords are generated after query rewriting the search words.

To achieve the above object, the present invention further provides a category prediction device, including:

A determining module, configured to determine the target keyword input by the user;

A matching module, configured to match the target keyword with the category correspondence to determine the predicted category;

Wherein, the category correspondence includes: the first category correspondence between keywords and categories determined through historical click behavior data, and/or the second category correspondence between commodity vocabulary and categories in the commodity library. A category correspondence, and/or a third category correspondence between vocabulary and categories of commodities that are successfully traded within a predetermined period.

To achieve the above object, the present invention further provides an electronic device, comprising:

memory for storing computer programs;

A processor, configured to implement the steps of the above category prediction method when executing the computer program.

To achieve the above object, the present invention further provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above category prediction method are realized.

It can be known from the above scheme that the embodiment of the present invention provides a category prediction method, device, equipment and storage medium; in this scheme, it is necessary to predetermine the category correspondence, and the category correspondence includes The first category correspondence between keywords and categories, and/or the second category correspondence between the vocabulary of commodities in the commodity library and the second category, and/or the vocabulary of commodities that are successfully traded within a predetermined period The third category correspondence between the category and the category; after obtaining the target keyword input by the user, the target keyword can be matched with the category correspondence to obtain the predicted category; it can be seen that this program is performing category When forecasting, not only can the category be predicted based on the first category correspondence determined by the historical click behavior data, but also the vocabulary of the commodity in the commodity library and the correspondence between the vocabulary and the category of the commodity that has been successfully traded within a predetermined period can be used to predict, Therefore, regardless of whether a click behavior occurs in the category, the category prediction can be accurately performed, thereby improving the accuracy of product recall.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only These are some embodiments of the present invention. Those skilled in the art can also obtain other drawings based on these drawings without creative work.

Fig. 1 is a schematic structural diagram of a category prediction device in a traditional scheme;

Fig. 2 is a schematic flow chart of a category prediction method disclosed in an embodiment of the present invention;

Fig. 3 is a schematic diagram of the corresponding relationship of the first category disclosed in the embodiment of the present invention;

Fig. 4 is a schematic diagram of the corresponding relationship of the second category disclosed in the embodiment of the present invention;

Fig. 5 is a schematic diagram of the overall flow of category prediction disclosed in the embodiment of the present invention;

FIG. 6 is a schematic structural diagram of a category prediction device disclosed in an embodiment of the present invention;

Fig. 7 is a schematic structural diagram of an electronic device disclosed by an embodiment of the present invention.

Detailed ways

It should be noted that, in order to clearly explain this scheme, the professional terms involved in this scheme are explained here:

Category: Classification of commodities refers to the collection of commodities within the scope of management in order to meet all or part of the needs of commodity production, circulation and consumption activities according to certain management purposes, and the selected basic characteristics of commodities are used as classification marks , successively summarized into several sub-aggregates (categories) with smaller scope and more consistent characteristics. On the e-commerce platform, the products will be neatly planned in a categorized manner, and when products need to be released, they need to be released in a level-by-level manner. The main categories include digital, home, office, clothing, beauty, transportation, etc. For example, computer products are classified under the [Digital] category. Computer chairs are classified under 【Office】category. Computer tablecloth products are classified under the [Home Furnishing] category.

Long tail words: In the search input keywords, the word length is relatively long. Generally, words with a length of more than 8 are called long-tail words; long-tail words, because of their longer length, contain more information and often have clearer goals.

Recall: Through the search keywords and long-tail words given by the user, go to the search engine to globally frame a batch of products that meet the user's filtering conditions.

See Figure 1, which is a schematic diagram of the structure of the category prediction device in the traditional solution. It can be seen from Figure 1 that in the traditional solution, the search click log can be obtained by searching for the click buried point, and the middle and high levels within a period of time can be selected from the search click log. Frequent query (search term) (for example, select a query whose search times are greater than or equal to 5 times in the past 15 days), and get the category prediction result of the query through the category prediction click model in the job, and store it in ES (ElasticSearch, database). When a user searches for a query, the category prediction result can be directly retrieved from ES through the category prediction interface and returned.

It can be seen that in the traditional way, the category prediction model records the high-frequency query hit by the search click into the offline dictionary, so the categories that have never been clicked and the categories with few clicks are not in the offline dictionary, so the search will not Return related categories. Therefore, in the embodiment of the present invention, a category prediction method, device, device, and storage medium are disclosed. This solution can realize category prediction corresponding to Top Query popular search words; it can also realize long-tail word prediction, and in the category Prediction is to consider user behavior preferences, thereby improving the accuracy of category prediction, thereby improving the recall rate of search.

The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some, not all, embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without creative efforts fall within the protection scope of the present invention.

Referring to FIG. 2 , a schematic flowchart of a category prediction method provided by an embodiment of the present invention includes:

S101. Determine the target keyword input by the user;

In this embodiment, when determining the target keyword input by the user, it is necessary to obtain the search term input by the user, and rewrite the search term to generate a plurality of target keywords.

Specifically, the search term is the content entered by the user when searching, and in order to improve the search quality, this solution rewrites the search term entered by the user to generate multiple target keywords, and each target keyword is related to the category match the corresponding relationship between items to obtain the predicted category corresponding to each target keyword; among them, the query rewrite (query rewrite, QR) technology can analyze and process the search words entered by the user to obtain the target that can better reflect the actual intention of the user. Keyword, the analysis process includes operations such as word segmentation and preprocessing on the search word. Therefore, in this solution, even if the search word entered by the user is a long-tail word, operations such as word segmentation can be performed on the long-tail word through query rewriting technology , so as to generate target keywords that are easier to predict categories, and improve the search hit rate of long-tail words.

S102. Match the target keyword with the category correspondence to determine the predicted category; wherein, the category correspondence includes: the first category correspondence between keywords and categories determined through historical click behavior data, and /or the second category correspondence between vocabulary and categories of commodities in the commodity library, and/or the third category correspondence between vocabulary and categories of commodities that are successfully traded within a predetermined period.

In this embodiment, it is necessary to establish a target correspondence in advance, and the target correspondence can be determined in various ways. Specifically, the first category correspondence between keywords and categories determined by historical click behavior data in this solution relationship, which can determine the corresponding relationship between the category where the click behavior occurs and the keyword. If the target keyword matches the keyword in the first corresponding relationship, then the category corresponding to the matched keyword is the target keyword. Predict categories; this program can determine the correspondence between categories and keywords with no click behavior or less click behavior through the second category correspondence between commodity vocabulary and categories in the commodity library. Mainly consider the correspondence between the vocabulary of the commodity and the category. If the target keyword matches the vocabulary of the commodity in the second correspondence, the category corresponding to the vocabulary of the matched commodity is the prediction of the target keyword. Category: This solution provides predictions for categories based on the positive correlation between the user's search query and the final traded commodity through the third category correspondence between the vocabulary of the successfully traded commodity within the predetermined period and the category .

To sum up, this program can predict categories from multiple perspectives, not only predicting categories based on the category correspondence determined by historical click behavior data, but also using the vocabulary of commodities in the commodity library and the success of transactions within a predetermined period. The prediction of the corresponding relationship between the vocabulary of the product and the category makes it possible to accurately predict the category regardless of whether the click behavior occurs in the category, thereby improving the accuracy of product recall.

Based on the above method embodiment, in this embodiment, a specific generation process of the first category correspondence relationship is disclosed, and the process mainly includes the following steps:

Obtain the historical click behavior data of each keyword under the first category of each recalled product;

Determine the number of clicks of each keyword under each first category according to the historical click behavior data, and the total number of clicks of each keyword;

Using the number of clicks of each keyword under each first category and the total number of clicks of each keyword, the first weight value of each keyword is calculated to generate a first category correspondence.

Specifically, this embodiment can generate the first category correspondence through the click mining module; when generating the first category correspondence, the history of each query under the category of each recalled product can be counted for a period of time through the click model The historical click behavior data in the website can determine the correlation between keywords and categories according to user behavior. For example: the keyword entered by the user is "television", and the first category of recalled products includes: "flat-screen TV", "television stand", "laser TV", etc., if the user clicks The category of "flat-panel TV" is identified, indicating that the category of "flat-panel TV" is related to TV sets, and at this time, the corresponding relationship between "television sets" and "flat-panel TVs" is established.

It should be noted that, in order to improve the quality of keywords, keyword condition preprocessing can be performed after obtaining historical click behavior data and before determining the number of clicks of each keyword under each first category. Including: deleting invalid keywords in historical click behavior data; the invalid keywords include: keywords with special characters and keywords with a length of 1. In addition, this solution can calculate the category relevance according to the number of clicks of each category, and this solution reflects the relevance through the weight value. Specifically, the method of calculating the weight value of query and category (cate_id) in this solution is the conditional probability, that is, the probability that the user clicks on the corresponding category given the query:

cate_score1=P(cate_id|query)=qc/qq;

Wherein, cate_score1 is the first weight value, qc indicates the number of clicks of the keyword query on the category cate_id, and qq indicates the total number of clicks of the keyword query.

Moreover, after calculating the first weight value of each keyword in this solution, each first weight value can be sorted according to the numerical value, and it can be detected whether the weight value of each keyword under each first category is less than a predetermined threshold; If yes, delete the corresponding relationship between the keywords whose weight value is less than the predetermined threshold and the first category. In this way, low-frequency query words can be excluded.

Referring to FIG. 3 , a schematic diagram of the first category correspondence provided by the embodiment of the present invention; it can be seen from this figure that the first category correspondence not only includes the correspondence between keywords and categories, but also includes the correspondence between keywords and categories. The weight value of the category, such as: the keyword is "television", then the category corresponding to "television" recorded in the first category correspondence includes: "flat-panel television" with the first weight value of 0.44, the first category A "television rack" with a weight value of 0.21, a "laser TV" with a first weight value of 0.12, and so on.

It should be noted that the above-mentioned process of analyzing, filtering and storing user search and click behavior data to obtain a <query,[cate_id,weight]> dictionary data can be understood as an offline dictionary construction process. After constructing offline dictionary data , it is necessary to persist this data and expose the interface for the user to query and use through the interface. Among them, the persistent layer storage uses elasticsearch.

Based on the above method embodiment, in this embodiment, a specific generation process of the second category correspondence relationship is disclosed, and the process mainly includes the following steps:

Obtain the first basic information of each commodity in the commodity library; the first basic information includes commodity titles and attributes;

splitting each first basic information into a plurality of first commodity words;

Determining the co-occurrence times that each first commodity vocabulary and each second category in each first basic information co-occur in the first basic information of the same commodity;

determining the total number of occurrences of each first commodity term;

The second weight value of each first commodity vocabulary under each second category is calculated by using the co-occurrence times and total occurrence times of each first commodity vocabulary and each second category to generate a second category correspondence.

Specifically, this embodiment can generate the second category correspondence through the text semantic mining module; wherein, the basic information of the commodity includes the commodity title and attributes, and the attributes include the model, brand, etc. of the commodity; if the commodity title and the attribute If it is split into a single commodity vocabulary, the correlation between each commodity vocabulary and category is different, or the categories related to each commodity vocabulary are limited and have different weights. Therefore, in this solution, the weight value of each commodity vocabulary and category can be calculated, and the value range is generally controlled at [0,1], 0 means irrelevant, and 1 means very relevant. This process is equivalent to feature selection, the feature is the commodity vocabulary, and the important commodity vocabulary is selected. The importance of commodity vocabulary is determined by the clustering of words in categories, that is, if commodity vocabulary only appears in a few categories, the more discriminative a word is, the higher its importance will be.

In this embodiment, the <commodity title word/brand model word, category> is paired, the co-occurrence relationship between the commodity vocabulary and the category is counted, and the method for calculating the correlation score between the vocabulary and the category usually has conditional probability, Pointwise mutual information (Pointwise Mutual Information, PMI), chi-square, tf-dc (Term Frequency-distribution concentration, word weight calculation method), etc., based on the actual data test results, this program adopts the conditional probability calculation method, that is, it is calculated at a given In the case of a commodity vocabulary (token), the probability of co-occurrence of the category (cate_id) and the vocabulary:

cate_score2=P(cate_id|token)=tc/tt;

Among them, cate_score2 is the second weight value, tc indicates the co-occurrence times of the first commodity vocabulary (token) and a certain category (cate_id), and tt indicates the total occurrence times of the token.

Referring to FIG. 4 , a schematic diagram of the corresponding relationship of the second category provided by the embodiment of the present invention; it can be seen from the figure that the corresponding relationship of the second category not only includes the corresponding relationship between each commodity vocabulary and the category, but also includes the commodity vocabulary and the corresponding relationship. The weight value of the category, for example: the first basic information is "Midea air conditioner KFR-72LW/DY-PA400", and the commodity vocabulary that splits the first basic information includes: "Midea", "air conditioner", "KFR-72LW ", "DY-PA400", here select "Midea" as the first commodity vocabulary, each first basic information includes the following second categories: "air conditioner", "rice cooker", "electric fan", "electric water heater" , "Electric Kettle", after counting the co-occurrence relationship between the product word "Midea" and the above-mentioned second category, the word "Midea" appeared 555 times in total, of which 85 times appeared together with the category "air conditioner", and it appeared with the category "Air Conditioner". The category of "rice cooker" appeared 59 times together, 42 times with the category of "electric fan", 36 times with the category of "electric water heater", 26 times with the category of "heater/heater", and 26 times with the category of " The "electric kettle" category appears 20 times together; the following results are calculated by the above method: "air conditioner": 0.15, "rice cooker": 0.11, "electric fan": 0.08, "electric water heater": 0.06, "heater/heating Appliance": 0.05, "Electric Kettle": 0.04.

It should be noted that this solution can build an online dictionary by establishing a second category correspondence when not enough search-click behavior data is collected. The online dictionary complements the model's ability to predict long-tail word scenarios. The online dictionary builds a set of prediction capabilities in the long-tail word scene by training a large number of existing product titles. The long-tail query word does not appear in the historical search click behavior data, but is input as a new word. The online dictionary performs word segmentation and preprocessing on the input long-tail word, and then inputs it into the model to obtain the category prediction result. In addition, in order to improve the overall response time of the prediction results in this solution, the dictionary is cached at the memory level, because the determined query words are the same as the predicted results under the determined model.

Based on the above method embodiment, in this embodiment, a specific generation process of the third category correspondence relationship is disclosed, and the process mainly includes the following steps:

Acquiring second basic information of commodities successfully traded within a predetermined period; the second basic information includes commodity titles and attributes;

splitting each second basic information into a plurality of second commodity words;

Determining the co-occurrence times that each second commodity vocabulary and each third category in each second basic information co-occur in the first basic information of the same commodity;

determining the total number of occurrences of each second commodity term;

The third weight value of each second commodity vocabulary under each third category is calculated by using the co-occurrence times and total occurrence times of each second commodity vocabulary and each third category to generate a third category correspondence.

Specifically, in this embodiment, the third category correspondence can be generated through the category fusion module; when generating the third category correspondence, considering that there is a positive correlation between the user's search query and the final transaction commodity, the most recent period The basic information involved in determining the corresponding relationship of the second category is called the first basic information, and the basic information involved in determining the corresponding relationship of the third category is called It is the second basic information, and the second basic information also includes product title and attribute, and the attribute includes information such as product brand and model. After the second basic information is determined, the second basic information can be split into multiple second commodity vocabulary. In order to distinguish, the commodity vocabulary split when determining the correspondence relationship of the second category is called the first commodity vocabulary. The commodity vocabulary split when determining the corresponding relationship of the third category is called the second commodity vocabulary; the category determined in the historical click behavior data is called the first category, and the category in the first basic information is called For the second category, the category in the second basic information is called the third category; the weight value in the first category correspondence is called the first weight value, and the weight value in the second category correspondence is called The value is called the second weight value, and the weight value in the third category correspondence is called the third weight value.

In this solution, when determining the third weight value of each second commodity vocabulary and the third category item, the conditional probability calculation method can be used, that is, to calculate the category (cate_id ) and the probability of co-occurrence of the second commodity vocabulary:

cate_score3=P(cate_id|token)=tc/tt

Among them, cate_score3 is the third weight value, tc indicates the co-occurrence times of the second commodity vocabulary (token) and a certain category (cate_id), and tt indicates the total occurrence times of the token.

Referring to Fig. 5 , a schematic diagram of the overall process flow of category prediction provided by the embodiment of the present invention; based on Fig. 5, it can be seen that the click mining module establishes the corresponding relationship between search keywords and categories based on click buried point data through offline calculation; text semantic mining Through online calculation, the module establishes the corresponding relationship between commodities and categories in the commodity library based on the commodity dimension table data; the category fusion module establishes the corresponding probability relationship between the commodity vocabulary and categories that have been successfully traded recently; after the user searches for query input, Prediction is performed through the category prediction interface. In the prediction, the target keyword is first determined by query rewriting, and then the category prediction result is obtained with the help of the category correspondence relationship established by the above modules. After returning the category prediction result, the predicted category will be Participate in searching for product recalls. It should be noted that, in this embodiment, after determining the predicted category through the category correspondence, the weight value of the predicted category can also be determined. Therefore, when returning the predicted result, the weight value of each predicted category and the specific pass The prediction results obtained from the corresponding relationship of a category are returned together, so as to be used as a reference in product recalls.

To sum up, it can be seen that this solution forms the corresponding relationship between keywords and categories in the offline dictionary by storing the high-frequency query records hit by search clicks and the categories corresponding to the queries, forming a positive cycle and improving the accuracy of category prediction. Accuracy rate; this solution sets up a text semantic mining module for long-tail words that are not in the offline dictionary, categories that have never been clicked, and categories that have few clicks. Through the text semantic mining module, long-tail words and categories that are not in the offline dictionary The keyword can also hit the corresponding category; this scheme integrates the click mining module and the text semantic mining module through the category fusion module, so that the recent transaction successful commodity vocabulary and the corresponding commodity category establish a corresponding matching relationship, and, here The corresponding relationship has the ability to gradually strengthen and evolve, and the matching relationship will be gradually strengthened with use. It can be seen that through this solution, the category prediction results can be matched respectively after the word segmentation of the query and returned by integrating the weight value, so that the category accuracy of product recall has been significantly improved.

The category prediction apparatus, equipment, and storage medium provided by the embodiments of the present invention are introduced below. The category prediction apparatus, equipment, and storage medium described below and the category prediction method described above may be referred to each other.

Referring to Figure 6, a schematic structural diagram of a category prediction device provided by an embodiment of the present invention, including:

Determining module 11, is used for determining the target keyword of user input;

Matching module 12, is used for matching described target keyword and category correspondence, determines prediction category;

Wherein, the category correspondence includes: the first category correspondence between keywords and categories determined through historical click behavior data, and/or the second category correspondence between commodity vocabulary and categories in the commodity library. A category correspondence, and/or a third category correspondence between vocabulary and categories of commodities that are successfully traded within a predetermined period.

In another embodiment of the present invention, the device further includes: a click mining module;

The click mining module includes:

The first acquisition unit is used to acquire the historical click behavior data of each keyword under the first category of each recalled commodity;

The first determination unit is configured to determine the number of clicks of each keyword under each first category and the total number of clicks of each keyword according to the historical click behavior data;

The first calculation unit is used to calculate the first weight value of each keyword by using the number of clicks of each keyword under each first category and the total number of clicks of each keyword, and generate the corresponding relationship of the first category .

In another embodiment of the present invention, the click mining module further includes:

The first deletion unit is configured to delete invalid keywords in the historical click behavior data; the invalid keywords include: keywords with special characters and keywords with a length of 1.

In another embodiment of the present invention, the click mining module further includes:

The detection unit is used to detect whether the weight value of each keyword under each first category is less than a predetermined threshold; if so, trigger the second deletion unit;

The second deleting unit is used to delete the corresponding relationship between keywords whose weight value is less than a predetermined threshold and the first category.

In another embodiment of the present invention, the device further includes: a text semantic mining module;

The text semantic mining module includes:

The second acquisition unit is used to acquire the first basic information of each commodity in the commodity library; the first basic information includes commodity titles and attributes;

A first splitting unit, configured to split each first basic information into a plurality of first commodity words;

The second determining unit is used to determine the number of co-occurrence times that each first commodity vocabulary and each second category in each first basic information co-occur in the first basic information of the same commodity;

a third determining unit, configured to determine the total number of occurrences of each first commodity vocabulary;

The second calculation unit is used to calculate the second weight value of each first commodity vocabulary under each second category by using the co-occurrence times and total occurrence times of each first commodity vocabulary and each second category, and generate Correspondence to the second category.

In another embodiment of the present invention, the device further includes: a category fusion module;

The category fusion module includes:

The third obtaining unit is used to obtain the second basic information of the commodities successfully traded within the predetermined period; the second basic information includes commodity titles and attributes;

a second splitting unit, configured to split each second basic information into a plurality of second commodity words;

The fourth determination unit is used to determine the co-occurrence times of each second commodity vocabulary and each third category in each second basic information in the first basic information of the same commodity;

A fifth determining unit, configured to determine the total number of occurrences of each second commodity vocabulary;

The third calculation unit is used to calculate the third weight value of each second commodity vocabulary under each third category by using the co-occurrence times and total occurrence times of each second commodity vocabulary and each third category, and generate Correspondence to the third category.

In another embodiment of the present invention, the determining module is specifically configured to: obtain a search word input by a user, and rewrite the search word to generate a plurality of target keywords.

Referring to FIG. 7 , a schematic structural diagram of an electronic device provided by an embodiment of the present invention includes:

Memory 21, used to store computer programs;

The processor 22 is configured to implement the steps of the category prediction method described in any method embodiment above when executing the computer program.

In this embodiment, the device may be a terminal device or a server, and the terminal device includes a PC (Personal Computer, personal computer), a smart phone, a tablet computer, a palmtop computer, a portable computer, and the like.

The device may include a memory 21 , a processor 22 and a bus 23 .

Wherein, the memory 21 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card type memory (for example, SD or DX memory, etc.), magnetic memory, magnetic disk, optical disk, etc. The storage 21 may be an internal storage unit of the device in some embodiments, such as a hard disk of the device. The memory 21 may also be an external storage device of the device in other embodiments, such as a plug-in hard disk equipped on the device, a smart memory card (SmartMedia Card, SMC), a secure digital (Secure Digital, SD) card, a flash memory card ( Flash Card), etc. Further, the memory 21 may also include both an internal storage unit of the device and an external storage device. The memory 21 can not only be used to store application software and various data installed in the device, such as program codes for executing the category prediction method, etc., but also can be used to temporarily store data that has been output or will be output.

In some embodiments, the processor 22 may be a central processing unit (Central Processing Unit, CPU), a controller, a microcontroller, a microprocessor, or other data processing chips for running program codes or processing stored in the memory 21. Data, such as program codes for executing category prediction methods, etc.

The bus 23 may be a peripheral component interconnect standard (PCI for short) bus or an extended industry standard architecture (EISA for short) bus or the like. The bus can be divided into address bus, data bus, control bus and so on. For ease of representation, only one thick line is used in FIG. 7 , but it does not mean that there is only one bus or one type of bus.

Further, the device can also include a network interface 24, and the network interface 24 can optionally include a wired interface and/or a wireless interface (such as a WI-FI interface, a bluetooth interface, etc.), which are usually used for communication between the device and other electronic devices Establish a communication connection.

Optionally, the device may further include a user interface 25, which may include a display (Display), an input unit such as a keyboard (Keyboard), and the optional user interface 25 may also include a standard wired interface or a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode, Organic Light-Emitting Diode) touch device, and the like. Wherein, the display may also be properly referred to as a display screen or a display unit, and is used for displaying information processed in the device and for displaying a visualized user interface.

FIG. 7 only shows a device with components 21-25. Those skilled in the art can understand that the structure shown in FIG. 7 does not constitute a limitation to the device, and may include fewer or more components than shown in the figure. Or combine certain components, or different component arrangements.

An embodiment of the present invention also provides a computer-readable storage medium, where a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the category prediction method described in any of the above method embodiments is implemented A step of.

Wherein, the storage medium may include: U disk, mobile hard disk, read-only memory (Read-Only Memory, ROM), random access memory (Random Access Memory, RAM), magnetic disk or optical disk, etc., which can store program codes. medium.

Each embodiment in this specification is described in a progressive manner, each embodiment focuses on the difference from other embodiments, and the same and similar parts of each embodiment can be referred to each other.

The above description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention will not be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims (10)

1. A category prediction method, comprising:

determining target keywords input by a user;

matching the corresponding relation between the target keywords and the categories, and determining predicted categories;

wherein, the category correspondence includes: and determining a first category corresponding relation between the keywords and the categories through the historical click behavior data, and/or determining a second category corresponding relation between the vocabulary and the categories of the commodities in the commodity library, and/or determining a third category corresponding relation between the vocabulary and the categories of the commodities which are successfully transacted in a preset period.

2. The category prediction method of claim 1, further comprising, prior to determining the target keyword input by the user:

acquiring historical click behavior data of each keyword under a first category of each recall commodity;

determining the clicking times of each keyword under each first category and the total clicking times of each keyword according to the historical clicking behavior data;

and calculating a first weight value of each keyword by using the clicking times of each keyword under each first category and the total clicking times of each keyword, and generating a first category corresponding relation.

3. The category prediction method of claim 2, wherein before determining the number of clicks under each first category for each keyword based on the historical click behavior data, further comprising:

deleting invalid keywords in the historical click behavior data; the invalid keywords include: keywords of special characters and keywords of length 1.

4. The category prediction method of claim 2, further comprising, after calculating the first weight value for each keyword:

detecting whether the weight value of each keyword under each first category is smaller than a preset threshold value;

if yes, deleting the keywords with the weight value smaller than the preset threshold value and the corresponding relation of the first category.

5. The category prediction method of claim 1, further comprising, prior to determining the target keyword input by the user:

acquiring first basic information of each commodity in a commodity library; the first basic information comprises commodity titles and attributes;

splitting each first basic information into a plurality of first commodity words;

determining the co-occurrence times of each first commodity vocabulary and each second category in each first basic information in the first basic information of the same commodity;

determining the total occurrence number of each first commodity vocabulary;

and calculating a second weight value of each first commodity vocabulary under each second category by using the co-occurrence times and the total occurrence times of each first commodity vocabulary and each second category, and generating a second category corresponding relation.

6. The category prediction method of claim 1, further comprising, prior to determining the target keyword input by the user:

acquiring second basic information of the commodity which is successfully transacted in a preset period; the second basic information comprises commodity titles and attributes;

splitting each second basic information into a plurality of second commodity words;

determining the co-occurrence times of each second commodity vocabulary and each third category in each second basic information in the first basic information of the same commodity;

determining the total occurrence number of each second commodity vocabulary;

and calculating a third weight value of each second commodity vocabulary under each third category by using the co-occurrence times and the total occurrence times of each second commodity vocabulary and each third category, and generating a third category corresponding relation.

7. The category prediction method according to any one of claims 1 to 6, characterized in that the determining a target keyword input by a user includes:

acquiring search words input by a user;

and searching and rewriting the search word to generate a plurality of target keywords.

8. A category prediction apparatus, comprising:

the determining module is used for determining target keywords input by a user;

the matching module is used for matching the corresponding relation between the target keywords and the categories and determining the predicted categories;

wherein, the category correspondence includes: and determining a first category corresponding relation between the keywords and the categories through the historical click behavior data, and/or determining a second category corresponding relation between the vocabulary and the categories of the commodities in the commodity library, and/or determining a third category corresponding relation between the vocabulary and the categories of the commodities which are successfully transacted in a preset period.

9. An electronic device, comprising:

a memory for storing a computer program;

a processor for implementing the steps of the category prediction method of any one of claims 1 to 7 when executing the computer program.

10. A computer readable storage medium, characterized in that the computer readable storage medium has stored thereon a computer program which, when executed by a processor, implements the steps of the category prediction method of any of claims 1 to 7.