WO2016095727A1 - 类目映射关系的建立方法与装置 - Google Patents
类目映射关系的建立方法与装置 Download PDFInfo
- Publication number
- WO2016095727A1 WO2016095727A1 PCT/CN2015/096728 CN2015096728W WO2016095727A1 WO 2016095727 A1 WO2016095727 A1 WO 2016095727A1 CN 2015096728 W CN2015096728 W CN 2015096728W WO 2016095727 A1 WO2016095727 A1 WO 2016095727A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- category
- mapping relationship
- target object
- value
- feature keyword
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Ceased
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
Definitions
- the embodiments of the present application relate to the field of information processing technologies, and in particular, to a method and an apparatus for establishing a category mapping relationship between a keyword and a category, and a category and a category.
- the electronic website divides a plurality of target objects into categories according to specific attribute requirements, and forms a category system, thereby classifying a plurality of target objects through the category system.
- the categories divided by the electronic websites may vary in number, level, name, and the like.
- the category mapping relationship Since there may be differences in the category system of different electronic websites, it is necessary to use the category mapping relationship when encountering the need to transfer the target object under one category of an electronic website to the category system of another electronic website. . For example, if the A-electronic website needs to transfer a target object under the category a to the B-electronic website, then it is necessary to determine which category the target object belongs to in the B-electronic website, and then the target object can be transferred to the determined one. Under the relevant category of the B electronic website. The process of determining the category of the target object under a certain category of the A-electronic website in the B-electronic website involves the establishment of the category mapping relationship.
- the method first selects a target object(s) of the category from a category to be mapped of an electronic website as a sample object, and then uses the feature information of the sample object as a keyword to find and select the feature in another electronic website. The target object that the information matches, and determines the category to which the found target object belongs. If the ratio of the target object included in the category to the number of pre-selected sample objects is greater than a preset value, then the two electrons A category mapping relationship is established between the above categories of the website.
- This method can establish a mapping relationship between categories of different electronic websites, so that when a target object under a certain category of an electronic website needs to be transferred to a category system of another website, the category can be directly followed.
- the mapping relationship is implemented.
- the establishment of the system is based on the matching of the target object feature information, and the matching is mainly text matching, and the text matching method only achieves matching in a literal sense, resulting in low accuracy of the established category mapping relationship.
- the embodiment of the present application provides a method for establishing a category mapping relationship between a keyword and a category, and a category and a category, and a corresponding device, so as to improve the accuracy of the category mapping relationship.
- Obtaining a feature keyword and acquiring, by the feature keyword, a target object that matches the keyword and operation behavior data that operates on the target object;
- the correspondence between the feature keyword and the category whose category value meets the preset condition is determined as the category mapping relationship between the feature keyword and the category.
- a category has multiple category values
- a plurality of category values are summed, and the result of the summation operation is used as the category value of the one category.
- the operation behavior of the operation on the target object includes at least two types of operation behaviors, and the target objects reflected by the different types of operation behaviors are different in degree of acceptance, and each type of operation behavior is allocated according to the degree of acceptance of the target object.
- v(c j ) is the category value of the jth category
- k i is the operational behavior data amount of the i-th type of operation behavior on the target object
- w i is the weight of the i-th type of operation behavior
- N is a natural number greater than or equal to 2.
- the method further includes:
- the amount of operational behavior data for the i-th type of operational behavior on the target object is corrected as follows:
- the amount of operational behavior data for the corrected i-th type of operational behavior For the time decay function of the i-th type of operational behavior, t is the length of time from the current time when the operational action of the i-th type occurs;
- the corrected operational behavior data amount is used to calculate the category value of the category to which the target object belongs.
- the method further includes:
- the correspondence between the feature keyword and the category to which the M target objects belong is determined as the category mapping relationship between the feature keyword and the category.
- the embodiment of the present application further provides a method for establishing a category mapping relationship between a category and a category, and the method includes:
- the correspondence relationship between the category corresponding to the feature keyword of the first category is determined as the category mapping relationship between the category and the category.
- determining the correspondence between the category of the first category and the category keyword as the category mapping relationship between the category and the category includes:
- the category values of the same category are summed, and the category values are the category values obtained in the foregoing method; the result of the summation operation is taken as the The final category value of the same category;
- determining the correspondence between the category of the first category and the category keyword as the category mapping relationship between the category and the category includes:
- the category values of the same category are summed, and the category values are the category values obtained in the foregoing method; the result of the summation operation is taken as the The final category value of the same category;
- P is a natural number greater than or equal to 1;
- the correspondence between the first category and the P categories is determined as the category mapping relationship between the category and the category.
- the embodiment of the present application further provides an apparatus for establishing a category mapping relationship between a feature keyword and a category.
- the device comprises: an obtaining unit, a first category determining unit, a category value calculating unit and a first mapping relationship determining unit, wherein:
- the acquiring unit is configured to acquire a feature keyword, and use the feature keyword to acquire a target object that matches the keyword and operation behavior data that operates on the target object;
- the first category determining unit is configured to determine a category to which each target object that matches the feature keyword belongs;
- the category value calculation unit is configured to calculate a category value of a category to which each target object belongs according to operation behavior data of the target object;
- the first mapping relationship determining unit is configured to determine a correspondence relationship between the feature keyword and the category whose category value meets the preset condition as a category mapping relationship between the feature keyword and the category.
- the apparatus further includes a summation unit configured to perform a summation operation on the plurality of category values when the category has a plurality of category values, and use the result of the summation operation as the category of the category value.
- a summation unit configured to perform a summation operation on the plurality of category values when the category has a plurality of category values, and use the result of the summation operation as the category of the category value.
- the operation behavior of the operation on the target object includes at least two types of operation behaviors, and the target object reflected by the different types of operation behaviors is different in degree
- the apparatus further includes: a weight distribution unit;
- the weight allocating unit is configured to perform each type of operation according to the degree to which the target object is accepted Assign behavior weights,
- the category value calculation unit is specifically configured to calculate a category value of a category to which each target object belongs according to the following formula:
- v(c j ) is the category value of the jth category
- k i is the operational behavior data amount of the i-th type of operation behavior on the target object
- w i is the weight of the i-th type of operation behavior
- N is a natural number greater than or equal to 2.
- the apparatus further includes: a data amount correction unit, and the operation behavior data amount for performing the i-th type of operation behavior on the target object is corrected as follows:
- the amount of operational behavior data for the corrected i-th type of operational behavior For the time decay function of the i-th type of operational behavior, t is the length of time from the current time when the operational action of the i-th type occurs;
- the category value calculation unit uses the corrected operation behavior data amount to calculate the category value of the category to which the target object belongs.
- the apparatus further includes: a determining unit, a selecting unit, a second category determining unit, and a second mapping relationship determining unit, wherein:
- the determining unit is configured to determine, after obtaining the operation behavior data that operates on the target object, whether the data amount of the operation behavior data is greater than a first preset threshold, and if yes, triggering the first category determination unit; , triggers the selection unit:
- the selecting unit is configured to select, from the target objects that match the feature keyword, the top M target objects with the highest matching degree;
- the second category determining unit is configured to determine a category to which each of the M target objects belongs;
- the second mapping relationship determining unit is configured to determine a correspondence relationship between the feature keyword and the category to which the M target objects belong as a category mapping relationship between the feature keyword and the category.
- the embodiment of the present application also provides a device for establishing a category mapping relationship between a category and a category.
- the device includes: a keyword determining unit, a category obtaining unit, and a third mapping relationship determining unit, wherein:
- the keyword determining unit is configured to determine each target in the first category of the first electronic website Each feature keyword of the object;
- the category obtaining unit is configured to use a feature keyword to search for a category mapping relationship between the feature keyword and the category established by the foregoing method, and obtain a category corresponding to the feature keyword;
- the third mapping relationship determining unit is configured to determine a correspondence relationship between the categories of the first category and the feature keywords as a category mapping relationship between the category and the category.
- the third mapping relationship determining unit includes a first summing subunit, a first sorting subunit, a second summing subunit, and a first mapping relationship determining subunit, wherein:
- the first summation subunit is configured to perform the summation operation on the category of the same category in the category corresponding to the feature keyword, and the category value is the category obtained in the foregoing method. a value; the result of the summation operation is taken as the final category value of the same class;
- the first sorting subunit is configured to sort each category according to a category value
- the second summation subunit is configured to perform a summation operation on the category values of the first L categories, and the result of the summation operation is greater than a second preset threshold, where the L is a natural number greater than or equal to 1;
- the first mapping relationship determining subunit is configured to determine a correspondence between the first category and the L categories as a category mapping relationship between the category and the category.
- the third mapping relationship determining unit includes a first summation subunit, a normalization subunit, a second sorting subunit, a third summation subunit, and a second mapping relationship determining subunit, wherein:
- the first summation subunit is configured to perform a summation operation on the category of the same category in the category corresponding to the feature keyword, and the category value is in the foregoing claims 1 to 5.
- the normalized subunit is configured to normalize each category value
- the second sorting subunit is configured to sort each category according to the normalized value of the normalized processing
- the third summation sub-unit is configured to perform a summation operation on the normalized values of the first P categories, and the result of the summation operation is greater than a third preset threshold, where P is a natural number greater than or equal to 1;
- the second mapping relationship determining subunit is configured to determine a correspondence between the first category and the P categories as a category mapping relationship between the category and the category.
- the manner of the embodiment of the present application establishes a category mapping relationship between the feature keyword and the category,
- the mapping relationship is based on the operation behavior data, and the operation behavior data can better reflect the user's tendency toward the target object, so that the search, comparison and the like based on the category mapping relationship are more accurate.
- the manner of the embodiment of the present application can establish a category mapping relationship between the category and the category, not only considering the text matching of the target object, but also considering the operation behavior data of the target object, because the operation behavior data can better reflect the user pair.
- the tendency of the target object, so that the category mapping relationship established according to the present application is more precise and more in line with the user's needs.
- FIG. 1 is a schematic flowchart diagram of an embodiment of a method for establishing a category mapping relationship between a feature keyword and a category of the present application
- FIG. 2 is a schematic flowchart of an embodiment of a method for establishing a category mapping relationship between a category and a category of the present application
- FIG. 3 is a structural block diagram of an embodiment of an apparatus for establishing a category mapping relationship between a feature keyword and a category of the present application
- FIG. 4 is a block diagram showing the composition of an embodiment of a category mapping relationship establishing device between a category and a category of the present application.
- the category mapping relationship of the embodiment is a mapping relationship between a feature keyword and a category of an electronic website. Relationships can be used in a variety of scenarios. For example, when you need to find a target object in a complex content of an electronic website, you can edit one or more Characterizing the feature keyword of the target object attribute to be searched, and then using the feature keyword as a search term to search for a mapping relationship between the feature keyword and the category, obtaining a category corresponding to the feature keyword, and then according to some sort The predetermined rule acquires all or part of the target objects in the category.
- the target objects have a larger probability of containing the target objects that the user really needs, so that the user can compare and identify each other and select the most suitable target objects.
- the predetermined rule here may be set according to the user's own needs. For example, all the target objects may be arranged in ascending order (descending order) according to a certain attribute value, and may be displayed, or only the degree of matching with the feature keyword may be displayed. The highest target object(s). The embodiment shown in Fig. 1 will be described in detail below.
- Step S11 acquiring a feature keyword, and using the feature keyword to acquire a target object that matches the keyword and operation behavior data that operates on the target object;
- the feature keyword can better reflect the feature of the target object, and in the electronic website, the target object matching the feature keyword can be found through the feature keyword.
- the feature keyword may be pre-defined by the administrator of the electronic website, or may be obtained by analyzing and processing the historical keyword input by the user of the electronic website.
- the feature keyword and the corresponding target object can be acquired by at least the following two exemplary ways.
- An exemplary way is: when the electronic website management maintainer links (or sets) the target object to the electronic website, select one or more words that can summarize certain aspects (attributes) of the target object according to the situation (for example, some The product name, material, plate type, etc.
- the selected word is used as the feature keyword, and stored in the database, so that the feature keywords can be directly obtained from the database and matched with Target object.
- Another exemplary way is that after entering the electronic website, the user inputs certain words in the search box provided by the electronic website, in order to search for the target object he needs through the search engine, in this case, by means of searching
- the engine system uses a search term obtained by processing the words input by the search engine system as a feature keyword, and uses the search result searched by the search term as a target object that matches the feature keyword.
- the operation behavior data of the target object may also be obtained from the search engine system or the electronic website.
- users can perform multiple types of operations on the target objects provided by the electronic website.
- the user's operation behavior can be expressed as a click browsing, collecting, adding a shopping cart, trading the product, etc., and the operation behavior of the product will be Electronic websites are recorded to form operational behavior data.
- These operational behavior data reflect the user's acceptance of a certain target object, and the target object that is not operated by the user (for example, an item that has not been clicked) is more likely to reflect the user's demand tendency.
- the operational behavior data may be derived from the user's operational behavior on the target object as described above, and may also be derived from various operational behaviors of the target object by the management and maintenance personnel of the electronic website, such as recommendations, ratings, and the like.
- which subject is the operation behavior, and what kind of behavior can be the type of the operation behavior, as long as it does not conflict with the purpose of the present invention, it can be used as the operation behavior data corresponding to the target object of the present application.
- Step S12 determining a category to which each target object that matches the feature keyword belongs
- the target object corresponding to a certain feature keyword and the operation behavior data corresponding to the target object can be obtained, and then, for each target object, it is determined which category the respective target object belongs to.
- the determined "category" may also include multiple. That is to say, one feature keyword may correspond to multiple categories.
- Step S13 calculating a category value of a category to which each target object belongs according to the operation behavior data of the target object;
- the category value After determining the target object's category, you can calculate the category value for each category. There are various ways to calculate the category value. However, no matter which way of calculating the category value, the calculated result should reflect the operation behavior data of the target object under the category. Specifically, The operational behavior data of the target object under the category and the category value calculated based on this can be expressed as a positive proportional relationship, that is, the larger the operation behavior data is (small), and the calculated category value is correspondingly higher. Large (small) can also be reflected in the inverse proportional relationship. The forward and reverse proportional relationship here will determine the comparison between the subsequent step category value and the preset threshold.
- the vector can be only a two-dimensional vector, that is, one dimension is each category (c1, c2..ck).
- the other dimension is the category value (v(c1), v(c2)...v(c)) corresponding to each category.
- Step S14 determining a correspondence relationship between the feature keyword and the category whose category value meets the preset condition as a category mapping relationship between the feature keyword and the category;
- the category mapping relationship established here can be expressed in mathematical expression: f1 (word, vec), f1 is the established category mapping relationship, word is the feature keyword, and vec is the aforementioned based on the category and category values. vector).
- the screening process can grasp the manner of screening and the degree of screening according to actual needs. For example, in order to establish a category mapping relationship between a feature keyword and a category of an electronic website as comprehensively and fully as possible, a method of comparing the calculated category value with a predetermined threshold may be used.
- the preset threshold may be lowered (increased) in order to obtain more eligible categories.
- the preset threshold can be increased (reduced), thereby obtaining fewer but more desirable categories.
- the "comparison” here is whether the comparison category value is greater than the preset value.
- the threshold is such that the correspondence between the feature keyword and those categories greater than the preset threshold is determined as the category mapping relationship between the feature keyword and the category.
- the "comparison” here is whether the comparison category value is smaller than the first preset threshold, thereby The correspondence between the categories smaller than the first preset threshold is determined as the category mapping relationship between the feature keyword and the category.
- the preset thresholds in the two cases may be the same or different, and different preset thresholds may better match the calculation methods of different category values.
- step S13 the calculation of the category value can be implemented in various specific ways.
- an exemplary calculation method is given below, which mainly considers two factors. : First, the type of operational behavior of the class's operational behavior; second, The extent to which different types of operational behavior affect the category value.
- the data generated by the operation behavior of the same operation type is The merge is a class, that is, a data corresponding to the type is counted. For example, for an item on an electronic website, "click to browse” is a type of operation behavior, and the counted number of times the item is clicked is the operation behavior data of the "click to browse” operation behavior, "collected goods” For another type of operation behavior, the counted number of times the item is collected is the operational behavior data of the operation behavior of "collected goods”.
- the corresponding data amount generated by the "collected goods” behavior can be set to have a greater influence on the category value than the "click to browse” behavior, similarly, “join the shopping cart” More than “collected goods”, “transactional goods” is larger than “joining shopping carts”.
- the category value can be calculated according to the following formula:
- v(c j ) is the category value of the jth category
- k i is the operational behavior data amount of the i-th type of operation behavior on the target object
- w i is the weight of the i-th type of operation behavior
- N is a natural number greater than or equal to 2.
- the calculated category value can better reflect the user's preference for the target object, if the relationship between the feature keyword and the category is established based on the category value.
- the category mapping relationship will enable the various applications based on this category mapping relationship to more accurately give the required categories and target objects, thereby improving the user experience.
- some details may be further optimized during the specific application process.
- three reference optimization methods for understanding the technical solutions are exemplarily described:
- a feature keyword may have multiple target objects matched to it, and a target object may also be divided into multiple categories, so that the "character keyword ⁇ target object ⁇ target object belongs to
- the "category" obtained by this "cluster" may be numerous and may have the same category.
- it is possible to compare the category values of each of the calculated categories with the preset thresholds without distinguishing the categories it is also possible to obtain a category that can be mapped to the feature keywords, but this may lead to the establishment of a category.
- the category mapping relationship is not accurate.
- the feature keyword A has two matching target objects b1 and b2.
- the target object b1 belongs to the category c1 and also belongs to the category c2.
- the target object b2 belongs to the categories c2 and c3.
- v b1 (c2) and v b2 (c2) are directly compared with the preset threshold, they may all be smaller than the preset threshold (the calculation method of the forward relationship is used here), thereby being excluded from
- the feature keyword A establishes a category other than the mapping relationship, but in fact, since v b1 (c2) and v b2 (c2) are the category values obtained for the category c2, the result after summing them should be The threshold is compared for comparison, so that it is possible to retain the category as a mapping relationship with the feature keyword A.
- the second example the influence of the time factor on the category value.
- the user's point of interest changes on the time axis, that is, the user has a "migration" characteristic on the operation behavior of the target object.
- the operation behavior data is far away from the current operation time of the user.
- there may be large deviations For example, a year ago, the "click-to-browse" behavior for a target object generated 500,000 data, and the “collected goods” behavior generated 200,000 data, while the current target audience for the same period of time
- the amount of "click-to-view” is 400,000, and the amount of "collected goods” is 250,000.
- the embodiment of the present application introduces a time decay function when using each operation behavior data, and corrects the operation behavior data according to the following manner, so as to use the corrected operation behavior data amount to perform the category value of the category to which the target object belongs.
- the amount of operational behavior data for the corrected i-th type of operational behavior For the time decay function of the i-th type of operational behavior, t is the length of time from the current time when the operational action of the i-th type occurs.
- ⁇ can select a value smaller than 1 and greater than 0 as needed, for example, 0.962.
- the third example the processing of less operational data.
- some (some) target objects may be “unattended” or have very little attention.
- the category value of each target object belongs to the category value can be calculated as described above. Then, based on the calculated category value, the category mapping relationship is established, but this may be “not worth the loss” and consume too many resources.
- an embodiment of the present application considers that the calculation of the category value is no longer performed in this case, but is compensated as follows:
- the correspondence between the feature keyword and the category to which the M target objects belong is determined as the category mapping relationship between the feature keyword and the category.
- the first preset threshold here can be controlled as needed, if the resource is the main goal and the accuracy is taken into account. Sex, you can set the threshold higher, on the contrary, set it lower.
- the above content details the process of establishing a class mapping relationship between feature keywords and categories, and the optimization process that may be performed based on various actual situations.
- this basically considers the category mapping relationship from the inside of an electronic website. If such a category mapping relationship is established, the user can retrieve a target object, especially a target object of the same type as the target object (ie, other target objects under the same category or target objects under the associated category).
- a target object of the same type as the target object ie, other target objects under the same category or target objects under the associated category.
- it provides more choices, which increases the possibility that the user finally gets the target object he needs, and improves the user experience (for example, can provide multiple target objects in quality, price, etc.) Horizontal comparison of aspects).
- category mapping relationship in addition to the category mapping relationship between the feature keywords of such an electronic website and the category system of the electronic website, there may be another type of category mapping relationship in reality, that is, a category system of an electronic website.
- a category mapping relationship with the category system of another electronic website is a one-sided perspective, and at least a mapping relationship to a "category" is established, which may be called a category mapping relationship.
- An application scenario is: a seller publishes a certain product on the electronic website A, and now needs to publish the product to the electronic website B, but since the electronic websites A and B may belong to different operators in maintenance management, With different category systems, the difference between the two category systems must be considered when transferring the goods from electronic website A to electronic website B. In this case, the category mapping relationship between the category and the category is involved. With the category mapping relationship, the product can be found on the electronic website B by the category of the product on the electronic website A. The above categories, in order to achieve the smooth release of goods on different electronic websites.
- Another application scenario is: a seller publishes a product on the B2C/C2C electronic website, but the seller may also purchase the purchase as a buyer on a B2B electronic website.
- the B2B electronic website needs to recommend B2B to the user. If the product is listed on the B2C/C2C electronic website and can calculate the corresponding category that may be of interest to the B2B electronic website, then the targeted recommendation can be made.
- the targeted recommendation can be made.
- To calculate the category that the user is interested in on the B2B electronic website it also involves the category mapping relationship between the category and the category.
- the embodiment of the present application provides a method for establishing a mapping relationship between a category and a category based on the category mapping relationship between the feature keyword and the category. Referring to Figure 2, the flow of the setup method is shown.
- Step S21 determining each feature keyword of each target object in the first category of the first electronic website
- the feature keyword can be expressed as the name, material, and shape of the target object.
- the first category is the category that needs to establish a category mapping relationship, that is, if it is necessary to establish a category mapping relationship between a category in the first electronic website and the category of the second electronic website, That is, the category is taken as the first category.
- the "first" and "second" in the first category, the first electronic website, and the second electronic website are merely given for convenience in calling different categories and websites, and do not represent sequential relationships.
- Step S22 Using the feature keyword, searching for the category mapping relationship between the feature keyword and the category established by the foregoing method in the second electronic website, and obtaining the category corresponding to the feature keyword;
- the feature keyword may be a search term, and the search operation is performed in the category mapping relationship in the second electronic website. Since the category mapping relationship of the second electronic website includes a correspondence between the feature keyword and the category of the second electronic website, the second electronic website corresponding to the feature keyword in the first electronic website can be found. Related categories.
- Step S23 determining a correspondence relationship between the categories of the first category and the feature keyword as a category mapping relationship between the category and the category;
- the relevant category is successfully found from the second electronic website, and then the first category can be associated with the related category of the found second electronic website.
- the correspondence relationship is the category mapping relationship between the category and the category.
- the mathematical expression is: f2 (item, vec), and item is the first category.
- the class in addition to considering the category value in the category mapping relationship between the feature keyword and the category, the class is not used between the category and the category.
- the mesh value that is, all the categories found can be regarded as the category corresponding to the first category.
- the rules are filtered, and only those categories that satisfy the condition are finally used as the category corresponding to the first category to establish a category mapping relationship between the category and the category.
- a screening method is to compare the various types of eye values of all the discovered categories with a preset threshold, and use a category that is greater than the preset threshold as a class that establishes a mapping relationship with the first category. Head.
- this "single category comparison" approach you can also perform a "summary” category comparison, and this "summary” approach can first sort the found categories according to the category value, and then From large to small (or small to large), the category values are accumulated until the sum of the sums is greater than (less than) a preset threshold (for example, the second preset threshold), at this time, the accumulated categories (such as , the first L) can be used as a category with a category mapping relationship with the first category.
- a preset threshold for example, the second preset threshold
- the above-mentioned "summary" category comparison method is to directly accumulate the category values, and there may be a variant, that is, the category values of the found categories are first normalized, and then Sorting the categories according to the normalized normalized values, and then performing a summation operation until the result of the summation is greater than a predetermined threshold (for example, a third preset threshold), at which time, the accumulation has been performed.
- a predetermined threshold for example, a third preset threshold
- the first category contains multiple target objects, and each target object may have multiple feature keywords, then the first category is At this level, there may be duplicates in the categories obtained according to each feature keyword.
- the feature keyword A in the first category corresponds to the category a1
- the feature keyword B in the first category may also correspond to the category a1
- the above-mentioned "single category comparison formula" or "sum type" can be calculated by using the final category value as long as the category value is utilized.
- the establishment of the above-mentioned category mapping relationship is based on the current target object under the category. If a new target object is added due to the update of the electronic website under a certain category, then the increase may be made first.
- the target object performs the calculation of the category value and the search work of the category corresponding to the feature keyword, and then accumulates the results into the previous result without all recalculation, thereby greatly reducing the calculation amount and improving the system. performance.
- the reason is similar, and the description will not be repeated.
- the embodiment of the present application further provides an apparatus for establishing a category mapping relationship.
- a block diagram showing the composition of a device for establishing a class mapping relationship between a feature keyword and a category is shown.
- the apparatus includes: an obtaining unit 31, a first category determining unit 32, a category value calculating unit 33, and a first mapping relationship determining unit 34, wherein:
- the obtaining unit 31 is configured to acquire a feature keyword, and use the feature keyword to acquire a target object that matches the keyword and operation behavior data that operates on the target object;
- a first category determining unit 32 configured to determine a category to which each target object that matches the feature keyword belongs
- a category value calculation unit 33 configured to calculate a category value of a category to which each target object belongs according to the operation behavior data of the target object;
- the first mapping relationship determining unit 34 is configured to determine a correspondence relationship between the feature keyword and the category whose category value meets the preset condition as a category mapping relationship between the feature keyword and the category.
- the device embodiment can achieve the same or similar technical effects as the foregoing method embodiments.
- the device can be specifically set on an electronic website to establish a category mapping relationship between the category system of the electronic website and the feature keywords, thereby facilitating operations such as querying, comparing, comparing, etc. of the electronic website, and facilitating the establishment of other electronic The category mapping relationship between the website and the electronic website.
- the device embodiment can also be improved in various aspects to obtain a better technical effect or to meet a specific need.
- the apparatus may further include a summation unit for performing a summation operation on the plurality of category values, and using the result of the summation operation as the category value of the category.
- the apparatus may further include: a weight assignment unit, configured to: Each type of operation behavior is assigned a weight according to the degree to which the target object is accepted, so that the category value calculation unit can calculate according to the following formula when calculating the category value of the category to which each target object belongs:
- v(c j ) is the category value of the j-th category
- k i is the operational behavior data amount of the i-th type of operation behavior on the target object
- w i is the weight of the i-th type operation behavior
- N is a natural number greater than or equal to 2.
- the above apparatus may further include a correction unit, and the amount of operation behavior data for performing the ith type of operation behavior on the target object is corrected as follows:
- the amount of operational behavior data for the corrected i-th type of operational behavior For the time decay function of the i-th type of operational behavior, t is the length of time from the current time when the operational action of the i-th type occurs;
- the category value calculation unit can use the corrected operational behavior data amount when calculating the category value of the category to which each target object belongs.
- the above apparatus embodiment may further include: a determining unit 35, a selecting unit 36, a second category determining unit 37, and a second mapping relationship determining unit 38, wherein:
- the determining unit 35 is configured to determine, after obtaining the operation behavior data that operates on the target object, whether the data amount of the operation behavior data is greater than a first preset threshold, and if yes, triggering the first category determination unit; if not, Then trigger the selection unit:
- a selecting unit 36 configured to select, from the target objects that match the feature keyword, the top M target objects with the highest matching degree
- a second category determining unit 37 configured to determine a category to which each of the M target objects belongs
- the second mapping relationship determining unit 38 is configured to determine a correspondence relationship between the feature keyword and the category to which the M target objects belong as a category mapping relationship between the feature keyword and the category.
- FIG. 4 shows an apparatus for establishing a category mapping relationship between a category and a category in the embodiment of the present application.
- the apparatus includes: a keyword determining unit 41, a category obtaining unit 42, and a third mapping relationship determining unit 43, wherein:
- a keyword determining unit 41 configured to determine each feature keyword of each target object in the first category of the first electronic website
- the category obtaining unit 42 is configured to use a feature keyword to search for a category mapping relationship between the feature keyword and the category established by the foregoing method according to the second electronic website, to obtain a category corresponding to the feature keyword;
- the third mapping relationship determining unit 43 is configured to determine a correspondence relationship between the categories of the first category and the feature keywords as a category mapping relationship between the category and the category.
- the third mapping relationship determining unit 43 may include a first summing subunit 431, a first sorting subunit 432, a second summing subunit 433, and a third mapping relationship determining subunit 434, where:
- the first summation sub-unit 431 is configured to perform the summation of the category values of the same category in the category corresponding to the feature keyword, and the category value is any of the foregoing claims 1 to 5. a category value obtained in a method; the result of the summation operation is taken as the final category value of the same category;
- a first sorting sub-unit 432 configured to sort each category according to a category value
- the second summation sub-unit 433 is configured to perform a summation operation on the category values of the first L categories, and the result of the summation operation is greater than a third preset threshold, where the L is a natural number greater than or equal to 1;
- the third mapping relationship determining sub-unit 434 is configured to determine a correspondence between the first category and the L categories as a category mapping relationship between the category and the category.
- composition structure of the foregoing third mapping relationship determining unit may further include: a first summation subunit, a normalization subunit, a second sorting subunit, a third summation subunit, and a second mapping relationship determining subunit, among them:
- the first summation subunit is configured to perform a summation operation on the category of the same category in the category corresponding to the feature keyword, and the category value is in the foregoing claims 1 to 5.
- the normalized subunit is configured to normalize each category value
- the second sorting subunit is configured to sort each category according to the normalized value of the normalized processing
- the third summation sub-unit is configured to perform a summation operation on the normalized values of the first P categories, and the result of the summation operation is greater than a fourth preset threshold, where P is a natural number greater than or equal to 1;
- the second mapping relationship determining subunit is configured to determine a correspondence between the first category and the P categories as a category mapping relationship between the category and the category.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
一种特征关键词与类目间、类目与类目间的类目映射关系的建立方法及其装置。特征关键词与类目间类目映射关系建立的方法包括：获取特征关键词，利用该特征关键词获取与该关键词匹配的目标对象以及对目标对象进行操作的操作行为数据（S11）；确定与该特征关键词匹配的每个目标对象所属的类目（S12）；根据目标对象的操作行为数据计算每个目标对象所属类目的类目值（S13）；将特征关键词与类目值符合预设条件的类目之间的对应关系确定为特征关键词与类目间的类目映射关系（S14）。能够提高类目映射关系的准确性。
Description
本申请的实施方式涉及信息处理技术领域，尤其涉及一种关键词与类目以及类目与类目之间的类目映射关系的建立方法与装置。
随着互联网技术的发展，电子网站上积聚的信息内容越来越多。这些信息内容可以被看成一个个目标对象，为了便于电子网站使用者(比如，电子网站管理、维护人员，电子网站的访问者等)对这些目标对象进行检索、浏览、收藏等操作，通常情况下，电子网站会将众多的目标对象按照特定属性要求划分类目，形成类目体系，从而通过类目体系来实现对众多的目标对象分门别类。根据电子网站以及目标对象的不同情况，各个电子网站划分出来的类目在数量、层级、名称等方面可能各不相同。
由于不同电子网站的类目体系可能存在差别，在遇到需要将一个电子网站某个类目下的目标对象转移到另一个电子网站的类目体系下等情况时，便需要利用类目映射关系。比如，A电子网站需要将a类目下的某个目标对象转移到B电子网站，那么就需要先确定该目标对象在B电子网站中属于哪个类目，然后才能将该目标对象转移到确定的B电子网站的相关类目之下。确定A电子网站某类目下的目标对象在B电子网站中的所属类目的过程，便涉及类目映射关系的建立问题。
在现有技术中，存在一种类目与类目之间的类目映射关系的方法。该方法先从一个电子网站的待映射类目中选择该类目下某个(些)目标对象为样本对象，然后以样本对象的特征信息为关键词，在另一个电子网站中查找与该特征信息匹配的目标对象，并确定查找到的目标对象所属的类目，如果该类目下包含的目标对象与预先选择的样本对象的数量之比大于某个预设值，则在这两个电子网站的上述类目之间建立类目映射关系。
这种方法能够建立不同电子网站的类目之间的映射关系，从而在需要将一个电子网站某个类目下的目标对象转移到另一个网站的类目体系中时，可以直接按照该类目映射关系实现转移。但是，由于这种类目映射关
系的建立是以目标对象特征信息的匹配为基础的，而这种匹配主要为文本匹配，文本匹配方式仅从字面意义上实现匹配，导致建立的类目映射关系准确性较低。
发明内容
为了解决上述问题，本申请实施方式提供了一种关键词与类目以及类目与类目间的类目映射关系的建立方法与相应的装置，以提高类目映射关系的准确性。
本申请实施方式提供的关键词与类目间的类目映射关系的建立方法包括：
获取特征关键词，利用该特征关键词获取与该关键词匹配的目标对象以及对目标对象进行操作的操作行为数据；
确定与该特征关键词匹配的每个目标对象所属的类目；
根据目标对象的操作行为数据计算每个目标对象所属类目的类目值；
将特征关键词与类目值符合预设条件的类目之间的对应关系确定为特征关键词与类目间的类目映射关系。
优选地，如果一个类目具有多个类目值，对多个类目值进行求和运算，将求和运算结果作为该一个类目的类目值。
优选地，对目标对象进行操作的操作行为包括至少两种类型的操作行为，不同类型的操作行为反映的目标对象被接受的程度不同，根据目标对象被接受的程度为每种类型的操作行为分配权重，所述根据目标对象的操作行为数据计算每个目标对象所属类目的类目值具体为按照如下公式计算每个目标对象所属的类目的类目值：
其中：v(cj)为第j个类目的类目值，ki为对目标对象进行第i种类型的操作行为的操作行为数据量，wi为第i种类型的操作行为的权重，N为大于或等于2的自然数。
优选地，所述方法还包括：
对目标对象进行第i种类型的操作行为的操作行为数据量按照如下方式进行修正：
其中：
为修正后的第i种类型的操作行为的操作行为数据量，
为第i种类型的操作行为的时间衰减函数，t为第i种类型的操作行为发生之时距离当前时间的时长；
将修正后的操作行为数据量用于计算目标对象所属类目的类目值。
优选地，所述方法还包括：
在获取到对目标对象进行操作的操作行为数据后，判断操作行为数据的数据量是否大于第一预设阈值，如果是，则执行确定与特征关键词匹配的每个目标对象所属的类目步骤；如果否，则：
从与特征关键词匹配的目标对象中选择匹配度最高的前M个目标对象；
确定M个目标对象各自所属的类目；
将特征关键词与M个目标对象所属的类目之间的对应关系确定为特征关键词与类目间的类目映射关系。
本申请实施例还提供了一种类目与类目间的类目映射关系的建立方法，该方法包括：
确定第一电子网站的第一类目下每个目标对象的每个特征关键词；
利用特征关键词，查找前述权利要求1至4中任何一种方法建立的特征关键词与类目间的类目映射关系，得到所述特征关键词对应的类目；
将第一类目与所述特征关键词对应的类目之间的对应关系确定为类目与类目间的类目映射关系。
优选地，将第一类目与所述特征关键词对应的类目之间的对应关系确定为类目与类目间的类目映射关系具体包括：
如果特征关键词对应的类目有相同的类目，对相同类目的类目值进行求和运算，所述类目值为前述方法中得到的类目值；将求和运算的结果作为该相同类目的最终类目值；
按照类目值大小对各个类目进行排序，对前L个类目的类目值进行求和运算，求和运算的结果大于第二预设阈值，所述L为大于等于1的自然数；
将第一类目与L个类目之间对应关系确定为类目与类目间的类目映
射关系。
优选地，将第一类目与所述特征关键词对应的类目之间的对应关系确定为类目与类目间的类目映射关系具体包括：
如果特征关键词对应的类目有相同的类目，对相同类目的类目值进行求和运算，所述类目值为前述方法中得到的类目值；将求和运算的结果作为该相同类目的最终类目值；
对各个类目值进行归一化处理；
按照归一化处理后的归一化值的大小对各个类目进行排序，对前P个类目的归一化值进行求和运算，求和运算的结果大于第三预设阈值，所述P为大于等于1的自然数；
将第一类目与P个类目之间对应关系确定为类目与类目间的类目映射关系。
本申请实施方式还提供了一种特征关键词与类目间的类目映射关系的建立装置。该装置包括：获取单元、第一类目确定单元、类目值计算单元和第一映射关系确定单元，其中：
所述获取单元，用于获取特征关键词，利用该特征关键词获取与该关键词匹配的目标对象以及对目标对象进行操作的操作行为数据；
所述第一类目确定单元，用于确定与该特征关键词匹配的每个目标对象所属的类目；
所述类目值计算单元，用于根据目标对象的操作行为数据计算每个目标对象所属类目的类目值；
所述第一映射关系确定单元，用于将特征关键词与类目值符合预设条件的类目之间的对应关系确定为特征关键词与类目间的类目映射关系。
优选地，所述装置还包括求和单元，用于在一个类目具有多个类目值时，对多个类目值进行求和运算，将求和运算结果作为该一个类目的类目值。
优选地，对目标对象进行操作的操作行为包括至少两种类型的操作行为，不同类型的操作行为反映的目标对象被接受的程度不同，则所述装置还包括：权重分配单元；
所述权重分配单元，用于根据目标对象被接受的程度为每种类型的操
作行为分配权重，
所述类目值计算单元，具体用于按照如下公式计算每个目标对象所属的类目的类目值：
其中：v(cj)为第j个类目的类目值，ki为对目标对象进行第i种类型的操作行为的操作行为数据量，wi为第i种类型的操作行为的权重，N为大于或等于2的自然数。
优选地，所述装置还包括：数据量修正单元，用于对目标对象进行第i种类型的操作行为的操作行为数据量按照如下方式进行修正：
其中：
为修正后的第i种类型的操作行为的操作行为数据量，
为第i种类型的操作行为的时间衰减函数，t为第i种类型的操作行为发生之时距离当前时间的时长；
所述类目值计算单元将修正后的操作行为数据量用于计算目标对象所属类目的类目值。
优选地，所述装置还包括：判断单元、选择单元、第二类目确定单元和第二映射关系确定单元，其中：
所述判断单元，用于在获取到对目标对象进行操作的操作行为数据后，判断操作行为数据的数据量是否大于第一预设阈值，如果是，则触发第一类目确定单元；如果否，则触发选择单元：
所述选择单元，用于从与特征关键词匹配的目标对象中选择匹配度最高的前M个目标对象；
所述第二类目确定单元，用于确定M个目标对象各自所属的类目；
所述第二映射关系确定单元，用于将特征关键词与M个目标对象所属的类目之间的对应关系确定为特征关键词与类目间的类目映射关系。
本申请实施方式还提供了一种类目与类目间的类目映射关系的建立装置。该装置包括：关键词确定单元、类目获取单元和第三映射关系确定单元，其中：
所述关键词确定单元，用于确定第一电子网站的第一类目下每个目标
对象的每个特征关键词；
所述类目获取单元，用于利用特征关键词，查找前述方法建立的特征关键词与类目间的类目映射关系，得到所述特征关键词对应的类目；
所述第三映射关系确定单元，用于将第一类目与所述特征关键词对应的类目之间的对应关系确定为类目与类目间的类目映射关系。
优选地，所述第三映射关系确定单元包括第一求和子单元、第一排序子单元、第二求和子单元和第一映射关系确定子单元，其中：
所述第一求和子单元，用于在特征关键词对应的类目有相同的类目，对相同类目的类目值进行求和运算，所述类目值为前述方法中得到的类目值；将求和运算的结果作为该相同类目的最终类目值；
所述第一排序子单元，用于按照类目值大小对各个类目进行排序；
所述第二求和子单元，用于对前L个类目的类目值进行求和运算，求和运算的结果大于第二预设阈值，所述L为大于等于1的自然数；
所述第一映射关系确定子单元，用于将第一类目与L个类目之间对应关系确定为类目与类目间的类目映射关系。
优选地，所述第三映射关系确定单元包括第一求和子单元、归一化子单元、第二排序子单元、第三求和子单元和第二映射关系确定子单元，其中：
所述第一求和子单元，用于在特征关键词对应的类目有相同的类目，对相同类目的类目值进行求和运算，所述类目值为前述权利要求1至5中任何一种方法中得到的类目值；将求和运算的结果作为该相同类目的最终类目值；
所述归一化子单元，用于对各个类目值进行归一化处理；
所述第二排序子单元，用于按照归一化处理后的归一化值的大小对各个类目进行排序；
所述第三求和子单元，用于对前P个类目的归一化值进行求和运算，求和运算的结果大于第三预设阈值，所述P为大于等于1的自然数；
所述第二映射关系确定子单元，用于将第一类目与P个类目之间对应关系确定为类目与类目间的类目映射关系。
本申请实施例的方式建立了特征关键词与类目之间的类目映射关系，
该映射关系以操作行为数据为基础进行，而操作行为数据更能反映用户对目标对象的倾向，从而使得基于这种类目映射关系的搜索、比较等更加准确。此外，本申请实施例的方式可以建立类目与类目之间的类目映射关系不仅考虑了目标对象的文本匹配，而且考虑了目标对象的操作行为数据，由于操作行为数据更能反映用户对目标对象的倾向，从而使得按照本申请建立的类目映射关系更加精准，更加符合用户需要。
通过参考附图阅读下文的详细描述，本发明示例性实施方式的上述以及其他目的、特征和优点将变得易于理解。在附图中，以示例性而非限制性的方式示出了本发明的若干实施方式，其中：
图1为本申请的特征关键词与类目间的类目映射关系建立方法的一个实施例的流程示意图；
图2为本申请的类目与类目间的类目映射关系建立方法的一个实施例的流程示意图；
图3为本申请的特征关键词与类目间的类目映射关系建立装置的一个实施例的组成框图；
图4本申请的类目与类目间的类目映射关系建立装置的一个实施例的组成框图。
下面将参考若干示例性实施方式来描述本发明的原理和精神。应当理解，给出这些实施方式仅仅是为了使本领域技术人员能够更好地理解进而实现本发明，而并非以任何方式限制本发明的范围。相反，提供这些实施方式是为了使本公开更加透彻和完整，并且能够将本公开的范围完整地传达给本领域的技术人员。
参见图1，该图示出了本申请的类目映射关系建立方法的一个实施例，该实施例的类目映射关系为特征关键词与电子网站的类目之间的映射关系，这种映射关系可以在多种场景下使用。比如，当需要在电子网站纷繁复杂的内容中查找某个(条)需要的目标对象时，可以编辑一个或多个能
够表征将要查找的目标对象属性的特征关键词，然后以该特征关键词为检索词，检索特征关键词与类目间的映射关系，获得与该特征关键词对应的类目，然后按照某种预定的规则获取该类目下的全部或部分目标对象，这些目标对象中有较大的概率包含有用户真正需要的目标对象，从而用户可以进行相互比较、甄别后选定最符合要求的目标对象。这里的预定规则可以根据用户自身的需要进行设定，比如，可以是将全部的目标对象依据某个属性值进行升序(降序)排列，并呈现出来，也可以是仅显示与特征关键词匹配程度最高的某个(些)目标对象。下面对图1所示的实施例进行详细介绍。
步骤S11：获取特征关键词，利用该特征关键词获取与该关键词匹配的目标对象以及对目标对象进行操作的操作行为数据；
特征关键词可以较好地反映目标对象的特征，在电子网站中，通过特征关键词能够查找到与该特征关键词匹配的目标对象。特征关键词可以是电子网站的管理者预先定义的，也可以是对电子网站的用户输入的历史关键词进行分析、加工得到的。与此对应地，可以通过至少如下两种示例性方式获取特征关键词以及对应的目标对象。一种示例性方式是：电子网站管理维护者在将目标对象链接(或设置)到电子网站时，根据情况选定一个或多个能够概括目标对象某方面特征(属性)的词(比如，某个商品的商品名称、材质、板型等)，将该选定的词作为特征关键词，并将其存储到数据库之中，这样，便可以直接从该数据库中获取特征关键词以及与之匹配的目标对象。另一种示例性方式是：用户在进入电子网站后，会在电子网站提供的搜索框中输入某些词语，以期望通过搜索引擎搜索到他需要的目标对象，这种情况下可以借助于搜索引擎系统，将搜索引擎系统对用户输入的词语进行加工分析得到的检索词作为特征关键词，将利用检索词搜索到的检索结果作为与该特征关键词匹配的目标对象。
在获取特征关键词和与特征关键词相匹配的目标对象后，还可以从搜索引擎系统或电子网站中获取对目标对象的操作行为数据。通常情况下，用户可以对电子网站提供的目标对象进行多种类型的操作。以电子商务网站展现的某个商品为例，用户的操作行为可以表现为对该商品进行点击浏览、收藏、加入购物车、交易该商品等行为，这些对商品的操作行为将被
电子网站记录下来，形成操作行为数据。这些操作行为数据反映出了用户对某个目标对象的接受程度，相对于那些未被用户操作的目标对象(比如，没有被点击的商品)，更能体现用户的需求倾向。此外，操作行为数据除如上所述可以来自于用户对目标对象的操作行为外，还可以来自电子网站的管理维护者对目标对象的各种操作行为，比如，推荐、置顶等行为。在本申请实施例中，操作行为究竟来自哪个主体，操作行为的类型究竟可以是哪些种类的行为，只要与本申请发明目的不冲突，均可以作为本申请的目标对象对应的操作行为数据。
步骤S12：确定与该特征关键词匹配的每个目标对象所属的类目；
按照前述步骤可以获得某个特征关键词对应的目标对象以及目标对象对应的操作行为数据，然后，针对每个目标对象，确定各个目标对象属于哪个类目。通常情况下，一个特征关键词对应的目标对象可能存在多个，从而使得确定出来的“类目”也可能包括多个。也就是说，一个特征关键词可能对应多个类目。
步骤S13：根据目标对象的操作行为数据计算每个目标对象所属类目的类目值；
在确定目标对象的类目后，可以计算每个类目的类目值。对于计算类目值的具体方式，可以存在多种，但是，无论哪种计算类目值的方式，计算出来的结果应当能够反映该类目之下的目标对象的操作行为数据情况，具体地，类目之下的目标对象的操作行为数据与基于此计算出来的类目值之间可以体现为一种正向比例关系，即操作行为数据越大(小)，计算得到的类目值相应越大(小)，也可以体现为反向比例关系，这里的正向、反向比例关系将决定后续步骤类目值与预设阀值之间的比较方式。为了便于体现类目与该类目的类目值之间的关系，可以以向量方式来呈现，该向量可以仅仅为两维向量，即一维为各个类目(c1、c2..ck)，另一维为各个类目对应的类目值(v(c1)、v(c2)…v(c))，简单地，可以用数学式表示如下：
步骤S14：将特征关键词与类目值符合预设条件的类目之间的对应关系确定为特征关键词与类目间的类目映射关系；
计算出各个类目的类目值之后，可以判断这些类目值是否符合预设条件，从而将那些符合预设条件的类目筛选出来，建立特征关键词与这些类目之间的类目映射关系(这里建立的类目映射关系可以用数学式表示为：f1(word，vec)，f1为建立的类目映射关系，word为特征关键词，vec为前述基于类目与类目值得到的向量)。该筛选过程可以根据实际需要来把握筛选的方式与筛选的程度。比如，为了尽可能全面、充分地建立某个特征关键词与电子网站的类目之间的类目映射关系，可以采取将计算出来的类目值与某个预设阈值比较的方式进行筛选。具体筛选时，判断该计算出来的类目值是否大于(或小于)预设阈值，为获得较多的符合条件的类目可以降低(提高)预设阈值。而为了尽可能精准、有针对性建立某个特征关键词与电子网站的类目之间的类目映射关系，则可以提高(降低)预设阈值，从而获得较少但较为满足需要的类目。需要注意的是，这里到底采取将类目值与预设阈值进行“大于”的比较，还是进行“小于”的比较，如前所述，取决于前一步骤中采取哪种方式来计算类目值。如果类目值是正向地反映类目之下的目标对象的操作行为数据，即操作行为数据越大，类目值也越大，那么，这里的“比较”是比较类目值是否大于预设阈值，从而将特征关键词与那些大于预设阈值的类目之间的对应关系确定为特征关键词与类目之间的类目映射关系。道理类似，在类目值是反向地反映类目之下的目标对象的操作行为数据，那么，这里的“比较”是比较类目值是否小于第一预设阈值，从而将特征关键词与那些小于第一预设阈值的类目之间的对应关系确定为特征关键词与类目之间的类目映射关系。当然，这两种情形下的预设阈值可以相同，也可以不同，不同的预设阈值更能体现与不同的类目值计算方式的匹配。
在步骤S13中，叙及可以采取多种具体方式实现类目值的计算，为了更明确地说明本申请的技术方案，下面给出一种示例性计算方式，这种计算方式主要考虑两个因素：一是对类目的操作行为的操作行为类型；二是
不同操作行为类型对类目值的影响程度。
通常情况下，对目标对象的操作存在多种操作方式，这些不同的操作方式可以划归为不同的操作行为类型，在进行操作行为数据的统计过程中，相同操作类型的操作行为产生的数据被归并为一类，即统计出一个与该类型对应的数据。比如，对于电子网站上的某个商品，“点击浏览”为一种操作行为类型，统计得到的该商品被点击的次数即为“点击浏览”这一操作行为的操作行为数据，“收藏商品”为另一种操作行为类型，统计得到的该商品被收藏的次数即为“收藏商品”这一操作行为的操作行为数据。
从上述对操作行为类型的划分来看，不同的操作行为类型反映出目标对象被接受的程度是不相同的。比如，“收藏商品”这一操作行为类型比“点击浏览”这一操作行为更体现了用户对目标对象的选择倾向。为此，在考虑操作行为类型基础上，还可以针对不同的操作行为类型分配不同的权重，分配权重的依据通常可以参考这些操作行为类型体现出来的用户对目标对象的接受程度。也就是说，在计算类目值过程中，可以将“收藏商品”行为产生的相应数据量对类目值的影响程度设置得比“点击浏览”行为更大，类似地，“加入购物车”比“收藏商品”更大，“交易商品”比“加入购物车”更大。
由此，可以按照下式来计算类目值：
其中：v(cj)为第j个类目的类目值，ki为对目标对象进行第i种类型的操作行为的操作行为数据量，wi为第i种类型的操作行为的权重，N为大于或等于2的自然数。
由上述计算类目值的公式可以看出，这种计算出来的类目值能够较好地体现用户对目标对象的偏好程度，如果基于该类目值来建立特征关键词与类目之间的类目映射关系，那么将使基于这种类目映射关系的各种应用场合更能精准地给出需要的类目以及目标对象，从而提升用户体验。但是，在具体应用过程中，可能还可以对一些细节作进一步优化。这里，为简便起见，示例性地叙述三种供理解技术方案的参考优化方式：
示例性之一：多类目求和运算。如前所述，一个特征关键词可能存在
多个目标对象与之匹配，而一个目标对象也可能被划分到多个类目之中，这样，在由“特征关键词→目标对象→目标对象所属的类目”这条“线索”获取到的“类目”可能数量众多，而且可能存在相同的类目。尽管可以不甄别类目是否出现重复，而只将计算得到的每个类目值与预设阈值进行比较，同样能够得到可与特征关键词建立映射关系的类目，但是，这样可能会导致建立的类目映射关系不准确。假设特征关键词A存在两个匹配的目标对象b1、b2，目标对象b1属于类目c1，也属于类目c2，目标对象b2属于类目c2和c3，对于类目c1、c3而言，均仅得到一个类目值v(c1)、v(c3)类目，那么可以直接将v(c1)、v(c3)与预设阈值进行比较，以判断它们是否能够作为与特征关键词A对应的类目。但对于类目c2而言，存在基于目标对象b1的操作行为数据计算得到的类目值vb1(c2)，以及基于目标对象b2的操作行为数据计算得出的类目值vb2(c2)，这时，如果直接将vb1(c2)、vb2(c2)单独与预设阈值进行比较，可能均小于预设阈值(本处采用正向关系的计算方式)，从而被排除在可与特征关键词A建立映射关系的类目之外，而实际上由于vb1(c2)、vb2(c2)均是针对类目c2得到的类目值，应该将它们求和之后的结果与预设阈值进行比较，从而有可能作为与特征关键词A建立映射关系的类目被保留下来。也就是说，在实际应用过程中，如果存在一个类目具有多个类目值时，可以对多个类目值进行求和运算，将求和运算的结果作为该类目的类目值。通过这看似简单的求和操作，可以使基于类目值建立起来的类目映射关系更加符合实际情况。
示例性之二：时间因素对类目值的影响。在现实中，用户的兴趣点在时间轴线上是变化的，即用户对目标对象的操作行为存在“迁移性”的特点，基于该特点，如果以相对于用户当前操作时间较远的操作行为数据来估计用户的当前行为，将可能出现较大的偏差。比如，在一年前，针对某个目标对象的“点击浏览”行为产生的数据量为50万、“收藏商品”行为产生的数据量为20万，而当前的一段时间内针对同样的目标对象的“点击浏览”量为40万，“收藏商品”量为25万，尽管表面上看“点击浏览”量出现了下滑，但“收藏商品”量却呈现上升趋势，这不是说明该目标对象越来越不被人接受(40+25＜50+20)，而是相反，说明该目标对象越来
越被人接受(25＞20)，由此可以看出，通常情况下越靠近当前时间的用户操作行为数据越能正确反映目标对象被接受的程度，进而计算得到的类目值也越可靠。也就是说，在计算类目值(考虑各个操作行为数据)时，不仅要横向考虑不同操作行为类型的各自权重，而且还应当考虑时间因素对操作行为数据有效性的影响，并认识到这种影响是随着时间轴向后以衰减方式来发挥影响的。为此，本申请实施方式在使用各操作行为数据时，引入时间衰减函数，按照如下的方式对操作行为数据进行修正，以便利用修正后的操作行为数据量进行目标对象所属类目的类目值的计算：
其中：
为修正后的第i种类型的操作行为的操作行为数据量，
为第i种类型的操作行为的时间衰减函数，t为第i种类型的操作行为发生之时距离当前时间的时长。这里的θ可以根据需要选取小于1且大于0的数值，比如，0.962。
示例性之三：操作行为数据量偏少的处理。在实际应用过程中，某个(些)目标对象可能“无人问津”或者关注量极少，这种情况下，尽管同样可以按照上述方式对每个目标对象所属的类目进行类目值计算，进而依据该计算得到的类目值进行类目映射关系的建立，但是，这样可能“得不偿失”，消耗过多的资源。为此，本申请的一种实施例考虑在这种情况下不再进行类目值的计算，而采用如下方式进行弥补：
在获取到对目标对象进行操作的操作行为数据后，判断操作行为数据的数据量是否大于第一预设阈值，如果是，则执行确定与特征关键词匹配的每个目标对象所属的类目步骤；如果否，则：
从与特征关键词匹配的目标对象中选择匹配度最高的前M个目标对象；
确定M个目标对象各自所属的类目；
将特征关键词与M个目标对象所属的类目之间的对应关系确定为特征关键词与类目间的类目映射关系。
这种弥补方式省去了对类目值的计算，而直接将与目标对象匹配有关类目(比如前M个)作为建立类目映射关系的类目。这里的第一预设阈值可以根据需要对其进行控制，如果以节约资源为主要目标并兼顾精准
性，可以将该阈值设置得高一些，相反，则设置得更低一些。
上述内容详细叙述了建立特征关键词与类目间的类目映射关系的过程以及基于各种实际情况可能进行的优化性处理。但是，这基本上是从一个电子网站的内部来考虑类目映射关系。如果建立了这种类目映射关系，便可以使用户检索某个目标对象，尤其是与该目标对象同类的目标对象(即相同类目之下的其他目标对象或关联类目之下的目标对象)时，能够更方便地检索，提供更多的选择余地，从而也就提高了用户最终获得其需要的目标对象的可能性，改善了用户体验(比如，可以提供多个目标对象在质量、价格等方面的横向比较)。实际上，除了这种电子网站的特征关键词与该电子网站的类目体系之间的类目映射关系外，现实中还可能存在另一类类目映射关系，即一个电子网站的类目体系与另一个电子网站的类目体系之间的类目映射关系。也就是说，本申请的类目映射关系是从单侧角度而言，至少建立了到“类目”的映射关系即可称为类目映射关系。
为便于对类目与类目间的类目映射关系的理解，下面举两个应用实例。一种应用场景是：一个卖家在电子网站A上发布了某个商品，现在需要将该商品发布到电子网站B上，但是，由于电子网站A、B可能属于不同的运营商在维护管理，因而具有不同的类目体系，那么将该商品由电子网站A转移到电子网站B上必须考虑这两个类目体系的差异。这种情形下，便涉及到类目与类目之间的类目映射关系，具有该类目映射关系，便可通过对该商品在电子网站A上的类目查找到该商品在电子网站B上的类目，从而实现商品在不同电子网站上的顺利发布。另一种应用场景是：一个卖家在B2C/C2C电子网站上进行商品发布，但是，该卖家还可能作为买家在一个B2B电子网站上采购进货，这时，B2B电子网站需要向该用户推荐B2B上的商品，如果根据该用户在B2C/C2C电子网站上发布的商品，能够计算得到其在B2B电子网站上可能感兴趣的对应类目，那么便可进行针对性的推荐。而要计算该用户在B2B电子网站上感兴趣的类目，也涉及到类目与类目之间的类目映射关系问题。为了适应这些现实需要，本申请实施方式在特征关键词与类目之间的类目映射关系基础之上，还提供了一种类目与类目映射关系的建立方法。参见图2，该图示出了该建立方法的流程。
步骤S21：确定第一电子网站的第一类目下每个目标对象的每个特征关键词；
如前所述，特征关键词可以表现为目标对象的名称、材质、板型等各方面的属性，在确定特征关键词时，如果出现“脏词”、无意义的修饰词，那么应当剔除掉，从而使最终得到的特征关键词为真正有效的词。这里的第一类目即是需要建立类目映射关系的类目，也就是，如果需要建立第一电子网站中的某个类目与第二电子网站的类目之间的类目映射关系，即将该类目作为第一类目。当然，这里第一类目、第一电子网站、第二电子网站中的“第一”、“第二”仅仅是便于称呼不同类目、网站的方便而给出的，并不代表顺序关系。
步骤S22：利用特征关键词，查找第二电子网站中依据前述方法建立的特征关键词与类目间的类目映射关系，得到所述特征关键词对应的类目；
在第一电子网站的某个类目下获得部分或者全部的目标对象的特征关键词之后，即可以该特征关键词为检索词，到第二电子网站中的类目映射关系中进行查找操作，由于第二电子网站的类目映射关系包含有特征关键词与该第二电子网站的类目之间的对应关系，从而可以查找到与第一电子网站中的特征关键词对应的第二电子网站的相关类目。
步骤S23：将第一类目与所述特征关键词对应的类目之间的对应关系确定为类目与类目间的类目映射关系；
这里借助于特征关键词的“桥梁作用”，顺利地从第二电子网站中查找到相关的类目，那么便可以将第一类目与查找到的第二电子网站的相关类目建立对应关系，该对应关系即是类目与类目间的类目映射关系，用数学式表示是：f2(item，vec)，item为第一类目。建立了这种类目与类目之间的类目映射关系的情况下，上述提及的两个应用场景的问题便可迎刃而解了。
在上述的类目与类目的映射关系建立过程中，除了在特征关键词与类目之间的类目映射关系中考虑类目值之外，类目与类目之间并未使用到类目值，也就是说，可以将查找到的全部类目作为与第一类目具有对应关系的类目。但是，在实际应用过程中，可能需要对查找到的全部类目按照某
种规则进行筛选，仅将那些满足条件的类目才最终作为与第一类目对应的类目，以建立起类目与类目之间的类目映射关系。比如，一种筛选方式是将查找到的全部类目的各类目值逐个与预设的某个阈值进行比较，将大于该预设阈值的类目作为与第一类目建立映射关系的类目。除这种“单个类目比较”的方式外，还可以进行“总和式”类目比较，而且，这种“总和式”方式可以先按照类目值大小对查找到的类目进行排序，然后从大到小(或者从小到大)进行类目值累加，直到累加后的总和大于(小于)某个预设阈值(比如，第二预设阈值)，这时，已经累加的类目(比如，前L个)即可作为与第一类目具有类目映射关系的类目。
当然，上述的“总和式”类目比较方式是直接将类目值进行累加的，还可能存在的一种变形方式，即先对查找到的类目的类目值进行归一化处理，然后按照归一化处理后的归一化值的大小进行类目排序，再进行求和运算，直至求和的结果大于某个预设阈值(比如，第三预设阈值)，这时，已经累加的项目(比如，前P个)即可作为与第一类目具有类目映射关系的类目。
在进行上述类目映射关系中，还可能存在这样的情形：由于第一类目之下包含多个目标对象，每个目标对象又可能具有多个特征关键词，那么站在“第一类目”这个层级上，依据各个特征关键词获得的类目可能存在重复。比如，第一类目下的特征关键词A对应类目a1，第一类目下的特征关键词B也可能对应类目a1，那么应当将针对特征关键词A对应的类目a1的类目值vA(a1)与针对特征关键词B对应的类目目a1的类目值vB(a1)进行求和运算，得到一个与类目a1对应的最终的类目值[v(a1)＝vA(a1)+vB(a1)]，进而，上述无论是“单个类目比较式”还是“总和式”，只要利用到类目值均可以利用该最终的类目值进行计算。
还需要说明的是上述类目映射关系的建立是以类目下的当前目标对象为基础的，如果某个类目下由于电子网站的更新，新增加了目标对象，那么可以先就该增加的目标对象进行类目值的计算，以及与特征关键词对应的类目的查找工作，然后再将这些结果累计到上一次结果中，而不需要全部重新计算，从而大大减少了计算量，提高系统性能。对于第二电子网站中出现目标对象的更新，道理类似，不再重复叙述。
上述内容详细说明了如何建立类目映射关系(包括特征关键词与类目之间的类目映射关系和类目与类目之间的类目映射关系)，与建立类目映射关系的方法相对应，本申请实施方式还提供了建立类目映射关系的装置。参见图3，该图示出了建立特征关键词与类目间的类目映射关系的建立装置的组成框图。该装置包括：获取单元31、第一类目确定单元32、类目值计算单元33和第一映射关系确定单元34，其中：
获取单元31，用于获取特征关键词，利用该特征关键词获取与该关键词匹配的目标对象以及对目标对象进行操作的操作行为数据；
第一类目确定单元32，用于确定与该特征关键词匹配的每个目标对象所属的类目；
类目值计算单元33，用于根据目标对象的操作行为数据计算每个目标对象所属类目的类目值；
第一映射关系确定单元34，用于将特征关键词与类目值符合预设条件的类目之间的对应关系确定为特征关键词与类目间的类目映射关系。
该装置实施例可以取得与前述的方法实施例相同或类似的技术效果。该装置可以具体设置在电子网站上，以建立本电子网站的类目体系与特征关键词的类目映射关系，从而方便基于本电子网站的查询、比较、对比等操作，也方便建立别的电子网站与本电子网站之间的类目映射关系。此外，在实际应用过程中，该装置实施例还可以进行多方面的改进，以获得更优的技术效果或满足某个特定需要。
比如，在一个类目具有多个类目值时，上述装置还可以包括求和单元，用于对多个类目值进行求和运算，将求和运算结果作为该一个类目的类目值。再比如，如果对目标对象进行操作的操作行为包括至少两种类型的操作行为，不同类型的操作行为反映的目标对象被接受的程度不同，则所述装置还可以包括：权重分配单元，用于根据目标对象被接受的程度为每种类型的操作行为分配权重，这样，类目值计算单元在计算每个目标对象所属的类目的类目值时便可以按照如下的公式进行计算：
其中：v(cj)为第j个类目的类目值，ki为对目标对象进行第i种类型
的操作行为的操作行为数据量，wi为第i种类型操作行为的权重，N为大于等于2的自然数。
还比如，上述装置还可以包括修正单元，用于对目标对象进行第i种类型的操作行为的操作行为数据量按照如下方式进行修正：
其中：
为修正后的第i种类型的操作行为的操作行为数据量，
为第i种类型的操作行为的时间衰减函数，t为第i种类型的操作行为发生之时距离当前时间的时长；
这样，类目值计算单元在计算每个目标对象所属类目的类目值时可以使用该修正后的操作行为数据量。
此外，上述装置实施例还可以包括：判断单元35、选择单元36、第二类目确定单元37和第二映射关系确定单元38，其中：
判断单元35，用于在获取到对目标对象进行操作的操作行为数据后，判断操作行为数据的数据量是否大于第一预设阈值，如果是，则触发第一类目确定单元；如果否，则触发选择单元：
选择单元36，用于从与特征关键词匹配的目标对象中选择匹配度最高的前M个目标对象；
第二类目确定单元37，用于确定M个目标对象各自所属的类目；
第二映射关系确定单元38，用于将特征关键词与M个目标对象所属的类目之间的对应关系确定为特征关键词与类目间的类目映射关系。
参见图4，该图示出了本申请实施例的一种类目与类目间的类目映射关系的建立装置。该装置包括：关键词确定单元41、类目获取单元42和第三映射关系确定单元43，其中：
关键词确定单元41，用于确定第一电子网站的第一类目下每个目标对象的每个特征关键词；
类目获取单元42，用于利用特征关键词，查找第二电子网站的依据前述方法建立的特征关键词与类目间的类目映射关系，得到所述特征关键词对应的类目；
第三映射关系确定单元43，用于将第一类目与所述特征关键词对应的类目之间的对应关系确定为类目与类目间的类目映射关系。
该装置的各个单元对应的功能具有不同的实现方式，那么其对应的内部结构也可能存在差别。比如，上述第三映射关系确定单元43可以包括第一求和子单元431、第一排序子单元432、第二求和子单元433和第三映射关系确定子单元434，其中：
第一求和子单元431，用于在特征关键词对应的类目有相同的类目，对相同类目的类目值进行求和运算，所述类目值为前述权利要求1至5中任何一种方法中得到的类目值；将求和运算的结果作为该相同类目的最终类目值；
第一排序子单元432，用于按照类目值大小对各个类目进行排序；
第二求和子单元433，用于对前L个类目的类目值进行求和运算，求和运算的结果大于第三预设阈值，所述L为大于等于1的自然数；
第三映射关系确定子单元434，用于将第一类目与L个类目之间对应关系确定为类目与类目间的类目映射关系。
此外，上述第三映射关系确定单元的另一种组成结构还可以包括第一求和子单元、归一化子单元、第二排序子单元、第三求和子单元和第二映射关系确定子单元，其中：
所述第一求和子单元，用于在特征关键词对应的类目有相同的类目，对相同类目的类目值进行求和运算，所述类目值为前述权利要求1至5中任何一种方法中得到的类目值；将求和运算的结果作为该相同类目的最终类目值；
所述归一化子单元，用于对各个类目值进行归一化处理；
所述第二排序子单元，用于按照归一化处理后的归一化值的大小对各个类目进行排序；
所述第三求和子单元，用于对前P个类目的归一化值进行求和运算，求和运算的结果大于第四预设阈值，所述P为大于等于1的自然数；
所述第二映射关系确定子单元，用于将第一类目与P个类目之间对应关系确定为类目与类目间的类目映射关系。
应当指出的是，上述优选实施方式不应视为对本发明的限制，本发明的保护范围应当以权利要求所限定的范围为准。对于本技术领域的普通技术人员来说，在不脱离本发明的精神和范围内，还可以做出若干改进和润
饰，这些改进和润饰也应视为本发明的保护范围。
Claims (16)
- 一种特征关键词与类目间的类目映射关系的建立方法，其特征在于，该方法包括：获取特征关键词，利用该特征关键词获取与该关键词匹配的目标对象以及对目标对象进行操作的操作行为数据；确定与该特征关键词匹配的每个目标对象所属的类目；根据目标对象的操作行为数据计算每个目标对象所属类目的类目值；将特征关键词与类目值符合预设条件的类目之间的对应关系确定为特征关键词与类目间的类目映射关系。
- 根据权利要求1所述的方法，其特征在于，如果一个类目具有多个类目值，对多个类目值进行求和运算，将求和运算结果作为该一个类目的类目值。
- 根据权利要求2所述的方法，其特征在于，对目标对象进行操作的操作行为包括至少两种类型的操作行为，不同类型的操作行为反映的目标对象被接受的程度不同，根据目标对象被接受的程度为每种类型的操作行为分配权重，所述根据目标对象的操作行为数据计算每个目标对象所属类目的类目值具体为按照如下公式计算每个目标对象所属的类目的类目值：
其中：v(cj)为第j个类目的类目值，ki为对目标对象进行第i种类型的操作行为的操作行为数据量，wi为第i种类型的操作行为的权重，N为大于或等于2的自然数。 - 根据权利要求3所述的方法，其特征在于，所述方法还包括：对目标对象进行第i种类型的操作行为的操作行为数据量按照如下方式进行修正：
其中：为修正后的第i种类型的操作行为的操作行为数据量，
为第i种类型的操作行为的时间衰减函数，t为第i种类型的操作行为发生之 时距离当前时间的时长；
将修正后的操作行为数据量用于计算目标对象所属类目的类目值。 - 根据权利要求1至4中任何一项所述的方法，其特征在于，所述方法还包括：在获取到对目标对象进行操作的操作行为数据后，判断操作行为数据的数据量是否大于第一预设阈值，如果是，则执行确定与特征关键词匹配的每个目标对象所属的类目步骤；如果否，则：从与特征关键词匹配的目标对象中选择匹配度最高的前M个目标对象；确定M个目标对象各自所属的类目；将特征关键词与M个目标对象所属的类目之间的对应关系确定为特征关键词与类目间的类目映射关系。
- 一种类目与类目间的类目映射关系的建立方法，其特征在于，该方法包括：确定第一电子网站的第一类目下每个目标对象的每个特征关键词；利用特征关键词，查找第二电子网站中依据前述权利要求1至4中任何一种方法建立的特征关键词与类目间的类目映射关系，得到所述特征关键词对应的类目；将第一类目与所述特征关键词对应的类目之间的对应关系确定为类目与类目间的类目映射关系。
- 根据权利要求6所述的方法，其特征在于，将第一类目与所述特征关键词对应的类目之间的对应关系确定为类目与类目间的类目映射关系具体包括：如果在第一类目下的特征关键词对应的类目中有相同的类目，对相同类目的类目值进行求和运算，所述类目值为前述权利要求1至4中任何一种方法中得到的类目值；将求和运算的结果作为该相同类目的最终类目值；按照类目值大小对各个类目进行排序，对前L个类目的类目值进行求和运算，求和运算的结果大于第二预设阈值，所述L为大于等于1的自然数；将第一类目与L个类目之间对应关系确定为类目与类目间的类目映射关系。
- 根据权利要求6所述的方法，其特征在于，将第一类目与所述特征关键词对应的类目之间的对应关系确定为类目与类目间的类目映射关系具体包括：如果特征关键词对应的类目有相同的类目，对相同类目的类目值进行求和运算，所述类目值为前述权利要求1至5中任何一种方法中得到的类目值；将求和运算的结果作为该相同类目的最终类目值；对各个类目值进行归一化处理；按照归一化处理后的归一化值的大小对各个类目进行排序，对前P个类目的归一化值进行求和运算，求和运算的结果大于第三预设阈值，所述P为大于等于1的自然数；将第一类目与P个类目之间对应关系确定为类目与类目间的类目映射关系。
- 一种特征关键词与类目间的类目映射关系的建立装置，其特征在于，该装置包括：获取单元、第一类目确定单元、类目值计算单元和第一映射关系确定单元，其中：所述获取单元，用于获取特征关键词，利用该特征关键词获取与该关键词匹配的目标对象以及对目标对象进行操作的操作行为数据；所述第一类目确定单元，用于确定与该特征关键词匹配的每个目标对象所属的类目；所述类目值计算单元，用于根据目标对象的操作行为数据计算每个目标对象所属类目的类目值；所述第一映射关系确定单元，用于将特征关键词与类目值符合预设条件的类目之间的对应关系确定为特征关键词与类目间的类目映射关系。
- 根据权利要求9所述的装置，其特征在于，所述装置还包括求和单元，用于在一个类目具有多个类目值时，对多个类目值进行求和运算，将求和运算结果作为该一个类目的类目值。
- 根据权利要求9所述的装置，其特征在于，对目标对象进行操作的操作行为包括至少两种类型的操作行为，不同类型的操作行为反映的目 标对象被接受的程度不同，则所述装置还包括：权重分配单元；所述权重分配单元，用于根据目标对象被接受的程度为每种类型的操作行为分配权重，所述类目值计算单元，具体用于按照如下公式计算每个目标对象所属的类目的类目值：
其中：v(cj)为第j个类目的类目值，ki为对目标对象进行第i种类型的操作行为的操作行为数据量，wi为第i种类型的操作行为的权重，N为大于或等于2的自然数。 - 根据权利要求11所述的装置，其特征在于，所述装置还包括：数据量修正单元，用于对目标对象进行第i种类型的操作行为的操作行为数据量按照如下方式进行修正：
其中：为修正后的第i种类型的操作行为的操作行为数据量，
为第i种类型的操作行为的时间衰减函数，t为第i种类型的操作行为发生之时距离当前时间的时长；
所述类目值计算单元将修正后的操作行为数据量用于计算目标对象所属类目的类目值。 - 根据权利要求9至12中任何一项所述的装置，其特征在于，所述装置还包括：判断单元、选择单元、第二类目确定单元和第二映射关系确定单元，其中：所述判断单元，用于在获取到对目标对象进行操作的操作行为数据后，判断操作行为数据的数据量是否大于第一预设阈值，如果是，则触发第一类目确定单元；如果否，则触发选择单元：所述选择单元，用于从与特征关键词匹配的目标对象中选择匹配度最高的前M个目标对象；所述第二类目确定单元，用于确定M个目标对象各自所属的类目；所述第二映射关系确定单元，用于将特征关键词与M个目标对象所属的类目之间的对应关系确定为特征关键词与类目间的类目映射关系。
- 一种类目与类目间的类目映射关系的建立装置，其特征在于，该装置包括：关键词确定单元、类目获取单元和第三映射关系确定单元，其中：所述关键词确定单元，用于确定第一电子网站的第一类目下每个目标对象的每个特征关键词；所述类目获取单元，用于利用特征关键词，查找前述权利要求1至4中任何一种方法建立的特征关键词与类目间的类目映射关系，得到所述特征关键词对应的类目；所述第三映射关系确定单元，用于将第一类目与所述特征关键词对应的类目之间的对应关系确定为类目与类目间的类目映射关系。
- 根据权利要求14所述的方法，其特征在于，所述第三映射关系确定单元包括第一求和子单元、第一排序子单元、第二求和子单元和第一映射关系确定子单元，其中：所述第一求和子单元，用于在特征关键词对应的类目有相同的类目，对相同类目的类目值进行求和运算，所述类目值为前述权利要求1至5中任何一种方法中得到的类目值；将求和运算的结果作为该相同类目的最终类目值；所述第一排序子单元，用于按照类目值大小对各个类目进行排序；所述第二求和子单元，用于对前L个类目的类目值进行求和运算，求和运算的结果大于第二预设阈值，所述L为大于等于1的自然数；所述第一映射关系确定子单元，用于将第一类目与L个类目之间对应关系确定为类目与类目间的类目映射关系。
- 根据权利要求14所述的装置，其特征在于，所述第三映射关系确定单元包括第一求和子单元、归一化子单元、第二排序子单元、第三求和子单元和第二映射关系确定子单元，其中：所述第一求和子单元，用于在特征关键词对应的类目有相同的类目，对相同类目的类目值进行求和运算，所述类目值为前述权利要求1至5中任何一种方法中得到的类目值；将求和运算的结果作为该相同类目的最终类目值；所述归一化子单元，用于对各个类目值进行归一化处理；所述第二排序子单元，用于按照归一化处理后的归一化值的大小对各个类目进行排序；所述第三求和子单元，用于对前P个类目的归一化值进行求和运算，求和运算的结果大于第三预设阈值，所述P为大于等于1的自然数；所述第二映射关系确定子单元，用于将第一类目与P个类目之间对应关系确定为类目与类目间的类目映射关系。
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410784706.1 | 2014-12-16 | ||
| CN201410784706.1A CN105786810B (zh) | 2014-12-16 | 2014-12-16 | 类目映射关系的建立方法与装置 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2016095727A1 true WO2016095727A1 (zh) | 2016-06-23 |
Family
ID=56125897
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/CN2015/096728 Ceased WO2016095727A1 (zh) | 2014-12-16 | 2015-12-08 | 类目映射关系的建立方法与装置 |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN105786810B (zh) |
| WO (1) | WO2016095727A1 (zh) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106294674A (zh) * | 2016-08-02 | 2017-01-04 | 郑州悉知信息科技股份有限公司 | 一种信息检测方法及装置 |
| CN108334539B (zh) * | 2017-12-15 | 2021-10-15 | 努比亚技术有限公司 | 对象推荐方法、移动终端及计算机可读存储介质 |
| CN109165240A (zh) * | 2018-07-12 | 2019-01-08 | 北京猫眼文化传媒有限公司 | 一种列表生成方法及装置 |
| CN110874386A (zh) * | 2018-08-30 | 2020-03-10 | 北京京东尚科信息技术有限公司 | 建立类目映射关系的方法和装置 |
| CN110502685A (zh) * | 2019-08-02 | 2019-11-26 | 阿里巴巴集团控股有限公司 | 一种基于搜索引擎的数据优化方法、装置及设备 |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103336796A (zh) * | 2013-06-09 | 2013-10-02 | 北京百度网讯科技有限公司 | 一种直接展示广告商品的方法及系统 |
| CN103425677A (zh) * | 2012-05-18 | 2013-12-04 | 阿里巴巴集团控股有限公司 | 关键词分类模型确定方法、关键词分类方法及装置 |
| CN103577423A (zh) * | 2012-07-23 | 2014-02-12 | 阿里巴巴集团控股有限公司 | 关键词分类方法及系统 |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7627548B2 (en) * | 2005-11-22 | 2009-12-01 | Google Inc. | Inferring search category synonyms from user logs |
| CN103034660B (zh) * | 2011-10-10 | 2016-09-28 | 阿里巴巴集团控股有限公司 | 信息提供方法、装置及系统 |
| CN103577473B (zh) * | 2012-08-03 | 2018-10-12 | 北京京东尚科信息技术有限公司 | 分类消岐方法、分类消岐装置及其系统 |
| CN103810208B (zh) * | 2012-11-14 | 2019-03-01 | 腾讯科技（深圳）有限公司 | 一种类目映射方法及装置 |
-
2014
- 2014-12-16 CN CN201410784706.1A patent/CN105786810B/zh active Active
-
2015
- 2015-12-08 WO PCT/CN2015/096728 patent/WO2016095727A1/zh not_active Ceased
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103425677A (zh) * | 2012-05-18 | 2013-12-04 | 阿里巴巴集团控股有限公司 | 关键词分类模型确定方法、关键词分类方法及装置 |
| CN103577423A (zh) * | 2012-07-23 | 2014-02-12 | 阿里巴巴集团控股有限公司 | 关键词分类方法及系统 |
| CN103336796A (zh) * | 2013-06-09 | 2013-10-02 | 北京百度网讯科技有限公司 | 一种直接展示广告商品的方法及系统 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105786810A (zh) | 2016-07-20 |
| CN105786810B (zh) | 2019-07-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102663627A (zh) | 个性化推荐方法 | |
| US11960471B2 (en) | Using lineage to infer data quality issues | |
| CN109033101B (zh) | 标签推荐方法及装置 | |
| Bagchi | Performance and quality assessment of similarity measures in collaborative filtering using mahout | |
| WO2020037931A1 (zh) | 项目推荐方法、装置、计算机设备及存储介质 | |
| WO2016161976A1 (zh) | 选择数据内容向终端推送的方法和装置 | |
| US8285721B2 (en) | Mapping item records to product records | |
| CN109063171B (zh) | 基于语义的资源匹配方法 | |
| CN107153656B (zh) | 一种信息搜索方法和装置 | |
| CN106776660A (zh) | 一种信息推荐方法及装置 | |
| US20160042428A1 (en) | Personalized product recommendation based on brand personality scale | |
| CN102339438A (zh) | 一种商品信息的网络发布方法、系统和设备 | |
| WO2016095727A1 (zh) | 类目映射关系的建立方法与装置 | |
| WO2015124024A1 (zh) | 一种提升信息的曝光率的方法和装置、确定搜索词的价值的方法和装置 | |
| US20100332564A1 (en) | Efficient Method for Clustering Nodes | |
| US20220138771A1 (en) | Method and apparatus for group filtered reports | |
| CN107292666A (zh) | 销售潜力判断方法及装置 | |
| US20240211977A1 (en) | System and method for correlating and enhancing data obtained from distributed sources in a network of distributed computer systems | |
| CN107896153A (zh) | 一种基于移动用户上网行为的流量套餐推荐方法及装置 | |
| CN111008331A (zh) | 门店端的展示方法、装置、电子设备及存储介质 | |
| CN114549125B (zh) | 物品推荐方法及装置、电子设备和计算机可读存储介质 | |
| CN114493132B (zh) | 资源分配方法、装置和电子设备 | |
| CN112529646A (zh) | 一种商品分类方法和装置 | |
| WO2017016403A1 (zh) | 确定业务对象品牌指数信息的方法及装置 | |
| CN111581517A (zh) | 对象推送方法、装置、电子设备及存储介质 |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 15869235 Country of ref document: EP Kind code of ref document: A1 |
|
| NENP | Non-entry into the national phase |
Ref country code: DE |
|
| 122 | Ep: pct application non-entry in european phase |
Ref document number: 15869235 Country of ref document: EP Kind code of ref document: A1 |