WO2019019554A1 - 一种推荐信息的获取方法及装置、电子设备 - Google Patents

一种推荐信息的获取方法及装置、电子设备 Download PDF

Info

Publication number
WO2019019554A1
WO2019019554A1 PCT/CN2017/120045 CN2017120045W WO2019019554A1 WO 2019019554 A1 WO2019019554 A1 WO 2019019554A1 CN 2017120045 W CN2017120045 W CN 2017120045W WO 2019019554 A1 WO2019019554 A1 WO 2019019554A1
Authority
WO
WIPO (PCT)
Prior art keywords
tag
label
pool
knowledge map
user
Prior art date
Application number
PCT/CN2017/120045
Other languages
English (en)
French (fr)
Inventor
汤彪
张弓
苏婧
郭皓洁
覃玉清
侯培旭
Original Assignee
北京三快在线科技有限公司
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 北京三快在线科技有限公司 filed Critical 北京三快在线科技有限公司
Priority to CA3066941A priority Critical patent/CA3066941A1/en
Priority to EP17918888.3A priority patent/EP3623956A4/en
Priority to JP2019569754A priority patent/JP7065122B2/ja
Priority to KR1020197036939A priority patent/KR102370408B1/ko
Publication of WO2019019554A1 publication Critical patent/WO2019019554A1/zh
Priority to US16/715,085 priority patent/US20200117675A1/en

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/903—Querying
    • G06F16/9032—Query formulation
    • G06F16/90324—Query formulation using system suggestions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24—Querying
    • G06F16/245—Query processing
    • G06F16/2457—Query processing with adaptation to user needs
    • G06F16/24578—Query processing with adaptation to user needs using ranking
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/901—Indexing; Data structures therefor; Storage structures
    • G06F16/9027—Trees
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9535—Search customisation based on user profiles and personalisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/02—Knowledge representation; Symbolic representation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
    • G06Q40/12—Accounting

Definitions

  • the present application relates to the field of computer technology, and in particular, to a method and device for acquiring recommendation information, and an electronic device.
  • the suggestion is intended to associate the input text of the user (in the search box), recommending the relevant matching prompt words for the user to select, to reduce the user input cost, and to control the quality of the Lenovo results to ensure the search engine is convenient. Understand and retrieve to enhance the user search experience.
  • the drop-down prompt usually performs prefix matching based on the text input by the user. For example, the recommendation probability of each recommendation information is calculated according to the weight of the input prefix and the weight of each pull-down prompt result under each weight, such as searching for "fire” suggest will give “hot pot”
  • the recommendation information is used as a prompt information for the user to select. Further, in order to expand the recommendation information extension provided to the user, category information related to the prompt information may be recommended for the user to select for each prompt information.
  • the system selects the category label of the prompt information "shirt dress” according to the preset label vocabulary, and then recommends the prompt information.
  • Category information such as: long paragraph, self-cultivation, Korean version, etc., the recommended information is relatively simple.
  • the application provides a method for obtaining recommendation information to enrich the obtained recommendation information as much as possible.
  • an embodiment of the present application provides a method for obtaining recommendation information, including:
  • Tag knowledge map is a knowledge map based on a core word and describing a multi-dimensional information of the core word by a tag
  • a predetermined number of tags are selected from the tag pool and recommended to the user.
  • the embodiment of the present application provides a device for acquiring recommendation information, including:
  • a prompt information obtaining module configured to obtain prompt information to be displayed
  • a tag pool construction module configured to construct a tag pool based on the prompt information and a preset tag knowledge map, wherein the tag knowledge map is a knowledge map that is based on a core word and describes a multi-dimensional information of the core word by a tag;
  • a recommendation module is configured to select a preset number of labels from the label pool to recommend to the user.
  • an embodiment of the present application further discloses an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor executes the computer program
  • the method for obtaining the recommendation information described in the embodiment of the present application is implemented.
  • the embodiment of the present application provides a computer readable storage medium, where the computer program is stored, and when the computer program is executed by the processor, the steps of obtaining the recommendation information disclosed in the embodiment of the present application are implemented.
  • the method for obtaining the recommendation information disclosed in the embodiment of the present application obtains the prompt information to be displayed, constructs a label pool based on the prompt information and the preset label knowledge map, and then selects a preset number of labels from the label pool. Recommended for users, enriching the information recommended to users. Moreover, by selecting the label in the obtained label pool according to the preset method, the label recommended to the user changes in real time, and the novelty of the recommendation information is enhanced.
  • FIG. 1 is a flowchart of a method for acquiring recommendation information according to an embodiment of the present application.
  • FIG. 2 is a flowchart of a method for acquiring recommendation information according to another embodiment of the present application.
  • FIG. 3 is a schematic diagram showing the structure of an apparatus for acquiring recommendation information according to an embodiment of the present application.
  • FIG. 4 is a schematic diagram showing the structure of an apparatus for acquiring recommendation information according to another embodiment of the present application.
  • FIG. 5 is a schematic diagram showing the structure of an apparatus for acquiring recommendation information according to still another embodiment of the present application.
  • the method for obtaining recommendation information described in the embodiment of the present application can be applied to the search recommendation field. For example, after the user inputs a query word through the application page, the search engine retrieves multiple search results based on the prefix search and displays it as a prompt message. For each prompt information, the recommendation system acquires the corresponding recommendation information by calling the recommendation information acquisition method disclosed in the present application, and displays it at an appropriate location.
  • the method for obtaining recommendation information disclosed in this embodiment includes: Step 100 to Step 120.
  • Step 100 Obtain prompt information to be displayed.
  • the search engine After the user inputs the search keyword, the search engine gives a prefix matching method and returns the corresponding search result for display to the user. Then, the recommended application selects the best preset number of search results as the prompt information according to the matching degree or click rate of the search result, so that the recommended application is displayed on the search page to the user.
  • the prompt information to be displayed to the user may be obtained by calling an interface provided by the search engine.
  • Step 110 Construct a tag pool based on the prompt information and a preset tag knowledge map.
  • the tag knowledge map is: a knowledge map in which the prompt information is a core word and the corresponding dimension information of the prompt information is described by a label.
  • the prompt information is typically obtained based on the user's search log and a click log based on the search results. For example, many users click on the "car beauty" search result after entering “car”, then “car beauty” will be used as a reminder when creating a tag knowledge map.
  • the tag knowledge map is a knowledge map in which the hint information is the core word and the corresponding dimension information of the hint information is described by the label.
  • the tag knowledge map includes multiple core words. Around each core word, a multi-dimensional knowledge system of the core word is constructed, and finally a clustered knowledge map is formed.
  • the label is used to describe the information of each dimension of the core word. For example, for the core word “car beauty”, the information of the category dimension can be described as “internal beauty”, “lacquer care”, information of its brand dimension. It can be described as “auto workshop” and “automobile billion” at the second speed.
  • the labels for the message “automotive beauty” include: “internal beauty”, “lacquer care”, “auto workshop”, “automobile billion”.
  • the knowledge map of each of the prompt information is constructed according to the presentation of the prompt information by the data structure and the hierarchical and category membership of each label.
  • the dimensions involved in the label include: category, sub-category, additional attributes, etc., and the additional attributes may be: business information, brand information, geographic information, product feature information, and the like.
  • the core word and the core word tag are usually determined by analyzing the platform data and combining specific business needs.
  • the tag knowledge map can be represented by a tree structure.
  • the root node of the tree structure is the core word of each cluster knowledge map, and the leaf nodes of the tree structure are labels of various dimensions of the core word.
  • Step 120 Select a preset number of labels from the label pool to recommend to the user.
  • a preset number of tags are generally selected from the tag pool and recommended to the user, for example, each of the prompt information displays three recommended tags.
  • a preset number of labels may be randomly selected and displayed to the user; a label with a high click rate may also be selected for display to the user; or, by using a confidence interval upper bound algorithm, the pre-selection is selected from the label pool.
  • the number of labels with the highest recommended value is recommended to the user.
  • the data may be clicked according to the history of the tag, or the recommended value of the tag may be calculated by combining the historical click data of the tag with the user characteristics of the user who accepts the recommendation.
  • the method for obtaining the recommendation information disclosed in the embodiment of the present application obtains the prompt information to be displayed, constructs a label pool based on the prompt information and the preset label knowledge map, and then selects a preset number of labels from the label pool. It is recommended to the user to solve the problem that the obtained recommendation information existing in the prior art is single and fixed with respect to the preset tag vocabulary. By recommending a tag pool based on a preset tag knowledge map, the information recommended to the user is enriched. By selecting the tags in the obtained tag pool according to the preset method, the tags recommended to the user are changed in real time, and the novelty of the recommendation information is enhanced.
  • the method for obtaining recommendation information disclosed in this embodiment includes: Step 200 to Step 230.
  • step 200 a tag knowledge map is constructed.
  • the prompt information is usually obtained based on the user's search log and the click log of the search results. For example, many users click on the "car beauty" search result after entering “car”, then "car beauty” will be used as a reminder when creating a tag knowledge map.
  • the tag knowledge map is a knowledge map in which the hint information is the core word and the corresponding dimension information of the hint information is described by the label.
  • the label is a word related to the core word, such as a word describing each dimension attribute of the core word.
  • the tag knowledge map includes multiple core words, labels around each core word, and the hierarchical relationship between the core words and the tags constitutes a multi-dimensional knowledge system of the core words, and finally forms a cluster-like knowledge map.
  • the label is used to describe the information of each dimension of the core word. For example, for the core word "car beauty”, the information of the category dimension can be described as "internal beauty", "lacquer care", information of its brand dimension.
  • the labels for the message “automotive beauty” include: “internal beauty”, “lacquer care”, “auto workshop”, “automobile billion”.
  • the knowledge map of each of the prompt information is constructed according to the presentation of the prompt information by the data structure and the hierarchical and category membership of each label.
  • the dimensions involved in the label include: category, sub-category, additional attributes, etc., and the additional attributes may be: business information, brand information, geographic information, product feature information, and the like.
  • the core word and the core word tag are usually determined by analyzing the platform data and combining specific business needs.
  • Building a tag knowledge map includes constructing a tag knowledge map based on structured point of interest data, and/or constructing a tag knowledge map based on user behavior logs.
  • the tag knowledge map may be constructed only according to the structured interest point data, or the tag knowledge map may be constructed only according to the user behavior log, and the tag knowledge map may be constructed according to both the structured point of interest data and the user behavior log.
  • a specific method of constructing a tag knowledge map includes: determining a core word and the core based on a point of interest name, a category name, and an additional attribute name in the structured point of interest data a tag of a word; a tree-like relationship of each tag according to a hierarchical relationship of the category system of the structured point of interest data, a hierarchical relationship of the additional attribute system, and a correspondence between the category system and the additional attribute system; wherein each of the tags The root node of the tree relationship is the core word of the cluster tag knowledge map, and the leaf node is the label of the corresponding level.
  • POI data Structured Point of Interest data
  • POI data is usually a search or recommendation platform that describes product information.
  • POI data usually includes: a category system and an additional attribute system.
  • the category system is to set different category labels for products according to business needs, and the hierarchical relationship of the categories.
  • the category label includes: the parent category name and the subcategory name; the category hierarchy includes: the category to which it belongs, the parent category of the category to which it belongs, and the subcategories to which the category belongs.
  • the category of the parent category includes: manicure, hairdressing, yoga, dance, tattoo, etc.; the hairdressing category is further divided into: sub-categories such as hair dyeing, haircutting, hair styling, etc. , in turn, constitutes a hierarchical category relationship.
  • a category system based on POI data can acquire a knowledge tag map of a certain number of commodities.
  • the additional attribute system is to set different additional attribute tags for the goods according to the business requirements, and the hierarchical relationship of the additional attributes. Take the “nail” product as an example. Among them, “nail” is the name of the point of interest, and its additional attributes may include: additional attribute labels such as painting and matte. The next level of the "painted" label includes: rich, fresh, desert and other attributes.
  • the category system complements the information of the additional attribute system to obtain more comprehensive data for constructing the tag knowledge spectrum.
  • Each POI data includes a plurality of fields, which respectively represent a POI name, a parent category, a sub-category, and attribute information.
  • the category name corresponding to the POI name the hierarchical relationship of each category name, the additional attribute name, and the hierarchical relationship of each attribute name can be obtained.
  • the category system and the additional attribute system have an associated relationship. For example, the category level of "nail" is the parent category, and the category level of "painting" is the subcategory.
  • a tag knowledge map can be constructed based on the above information extracted from the POI data.
  • the core word is “nail”, and the lower-level labels include: “painting” and “matte”, according to which, the label knowledge map of “manicure” can be constructed. If the tag knowledge map is represented by a tree structure, the root node is “manicure”, and the leaf nodes include: “painting” and "matte”. In the specific implementation, it is also possible to further describe the additional attribute labels such as rich, fresh, and desert as the leaf nodes of the leaf node, "painting”, and improve the label knowledge map.
  • a specific method for constructing a tag knowledge map includes: determining a core word and a label of the core word based on a frequent item set of the mining user behavior log; Presetting the association relationship, establishing a tree relationship of the corresponding label; wherein the root node of the tree relationship is a core word of the cluster label knowledge map, and the leaf node is a label; the preset association relationship includes: a commodity and a merchant Relationship, relationship between business and business circle.
  • the application on the platform has a large number of user behavior logs, and by analyzing the historical behavior data of the massive users, the core words and the core word related tags can be obtained.
  • the mining algorithm is used to mine the frequent itemsets in the log data, and the core words and the labels of the core words are selected from the frequently mined items, and the search words can be used as the core words.
  • the core word tag is selected from the frequently mined items, and the hierarchical relationship of the core word to the tag is established.
  • “Shanghai Xiaopang” and “Shanghai Xiaopeng” constitutes the label of crayfish. If “crayfish” is used as the root node of the tree-like relationship, “Shanghai Xiaopang” is used as the leaf node of the tree-like relationship, which constitutes a part of the label knowledge map of the core word "crayfish”.
  • the distributed business circle acts as a label for the merchant, establishing a hierarchical relationship from the merchant to the business circle, and further constructing a tag knowledge map.
  • the merchant is regarded as the root node of the tree relationship, and the business circle is used as the leaf node of the tree relationship.
  • the core words and labels can be selected according to the business requirements, and the label knowledge map of the label core words is further established.
  • the preset association relationship can also be set to other relationships according to specific business requirements, which are not enumerated here.
  • the merchants with high quality and high click rate and the high quality business circle are recommended to the user, which further enhances the user experience.
  • Step 210 Obtain prompt information to be displayed.
  • the search engine gives a prefix matching method and returns the corresponding search result for display to the user. Then, the recommended application selects the best preset number of search results as the prompt information according to the matching degree or click rate of the search result, so that the recommended application is displayed on the search page to the user.
  • the prompt information to be displayed to the user may be obtained by calling an interface provided by the search engine. In a specific implementation, the prompt information may be displayed to the user in the form of a drop-down prompt word, or may be displayed to the user in other forms.
  • Step 220 Construct a tag pool based on the prompt information and a preset tag knowledge map.
  • the tag knowledge map is: a knowledge map in which the prompt information is a core word and the corresponding dimension information of the prompt information is described by a label.
  • the constructing the label pool based on the prompt information and the preset label knowledge map comprises: matching the prompt information with each core word in the preset label knowledge map; All tags in the tag knowledge map corresponding to the word are added to the tag pool.
  • the tag knowledge map is represented by a tree structure as an example.
  • the root node of the tree structure is the core word of each cluster knowledge map, and the leaf nodes of the tree structure are labels of various dimensions of the core word.
  • the number of labels corresponding to each prompt information is related to the level of the knowledge map of the prompt information. The higher the level, the more labels there are.
  • the recommended words displayed to the user for each prompt information that is, the number of tags is limited. For example, limit each drop-down word to display up to three related tags, while limiting the entire screen, allowing up to nine labels to be displayed to avoid flooding the screen labels. Therefore, in the order of the highest level of the knowledge map hierarchy, the tag construction tag pool in the highest layer of the knowledge map in which the prompt information matches the successful core word is taken.
  • Step 230 Select a preset number of labels from the label pool to recommend to the user.
  • a preset number of tags are generally selected from the tag pool and recommended to the user, for example, each of the prompt information displays three recommended tags.
  • a preset number of labels can be randomly selected for display to the user, but the method of selecting will cause many unpopular labels to be overexposed, wasting valuable display opportunities; and can also select a high click rate.
  • the label is displayed to the user.
  • the drawback of this method is that the labels displayed to each user are the same, and a large number of labels have no display opportunities, which impairs the user experience.
  • the selecting a preset number of labels from the label pool is recommended to the user, and: selecting, by using a confidence interval upper bound algorithm, a preset number of labels from the label pool to recommend to the user.
  • Determining, by the confidence interval upper bound algorithm, a preset number of labels from the label pool to the user further comprising: estimating, according to historical behavior data of the label, the expected revenue of each label in the label pool; The total number of times the history tag of the prompt information is displayed and the total number of times each tag in the tag pool is displayed, respectively determining a revenue adjustment indicator of each tag in the tag pool; and the expected revenue and the revenue adjustment indicator And the recommended value of each label in the label pool; selecting a preset number of labels with the highest recommended value from the label pool is recommended to the user.
  • the expected revenue of the tag revenue can be expressed as:
  • the estimating the expected benefit of each tag in the tag pool according to the historical behavior data of the tag by the user is implemented by two methods.
  • the expected revenue of the corresponding tag in the tag pool is estimated based on the historical click rate of the same prompt information as the tag.
  • the prompt information display position and the label display position are usually set.
  • the prompt information display position is used to display the prompt information searched according to the search keyword, and the following prompt message is used;
  • the label display position is used to display the label of the prompt information, and the category of the prompt word is as follows.
  • the expected revenue of the corresponding label in the label pool may be estimated according to the historical click rate after the label is displayed as the prompt information.
  • the historical click rate of the prompt information refer to any technique well known to those skilled in the art.
  • the prompt information returned by the search engine includes: “car” and “car beauty” as an example. If the recommended application identifies the label for the prompt message “Car Beauty”: “Automobile Yiyi”, “Auto Workshop”, “Automobile Station” and other labels, it is assumed that the “Car Yiyi” has not been labeled as a reminder before. Recommend to the user, according to the historical data of the search engine, determine the historical click rate of the prompt information "Automotive billion”, and then determine the expected return of the label "Automotive billion” according to the historical click rate of the prompt information "Automotive billion”.
  • the historical click rate of the prompt information "Automotive billion” is used as the expected return of the label "Automotive billion", or the historical click rate of the prompt information "Automotive billion” is multiplied by a coefficient, and then the product is used as a label.
  • the expected return of the car billion when the expected revenue of the corresponding label in the label pool is estimated, for example, t in the formula 1 may be the label “Automotive Beauty” label “Automotive”
  • Other calculation methods may also be adopted according to specific business requirements, and the specific calculation method is not limited in this application.
  • the expected revenue of the corresponding tag in the tag pool is estimated.
  • the expected revenue of the corresponding tag in the tag pool may be estimated according to the current user feature and the estimated click rate of the tag.
  • the historical click rate data of the label will be updated in real time, and further calculating the expected return of the label based on the updated historical click rate data will improve the accuracy of the calculation result.
  • the logistic regression algorithm (LR, Logistic Regression) can be used to train the user click probability model to more accurately calculate the current expected return of the user.
  • LR Logistic Regression
  • ⁇ is the feature vector of the feature of the user u and the estimated click rate feature of the tag j of the prompt information X.
  • the user characteristics include, but are not limited to, one or more of a user searching for a geographical location, a search period, a category preference of the user for the label, a user's preference for the recommended merchant, and the like.
  • the estimated clickthrough rate feature can be the historical clicks of tag j or the estimated click rate based on user behavior data.
  • W is a parameter vector of the logistic regression algorithm LR, which corresponds to the feature vector of the user, and W is specifically the weight of each dimension feature in the feature vector ⁇ .
  • T is a transposed matrix corresponding to ⁇ and W.
  • the model W vector can be trained by using the open source LR algorithm package according to the historical click data of the user u on the label, that is, the behavior data of whether the label j is clicked. Then, the user's expected benefit is estimated in conjunction with the feature vector ⁇ of the user u, the targeted recommendation of the user is achieved, and a variety of recommended tags are provided as the click behavior changes.
  • the feature vector ⁇ of the user refer to any technique known to those skilled in the art, and details are not described herein again.
  • the feature vector of the user includes at least a click rate feature.
  • the revenue adjustment indicator bonus of each label is used to balance the display probability of the label according to the needs of the business.
  • the revenue adjustment indicator of each label in the label pool is determined according to the total number of times the history label is displayed and the total number of times each label in the label pool is displayed.
  • the income adjustment indicator bonus may be calculated using Equation 3 below.
  • the impression will be obtained, and the recommended value will be appropriately increased.
  • the tag assigns some impressions.
  • the inverse adjustment function of other expressed t and T j,t can also be used to calculate the income adjustment index, which is not enumerated here.
  • a preset number of labels with the highest recommended value is selected from the label pool and recommended to the user. Take the preset quantity equal to 2 as an example. Assume that the label determined by the “Car Beauty” message includes: “Car Yiyi”, “Auto Workshop”, “Car Station” and other labels. After calculation, select the label with the highest recommended value. The car Yiyi” and “Auto Workshop” are recommended to users.
  • the method for obtaining the recommendation information disclosed in the embodiment of the present application after constructing the label knowledge map in advance, and after obtaining the prompt information to be displayed, constructing a label pool based on the prompt information and the preset label knowledge map, and then A preset number of tags are selected in the tag pool and recommended to the user, enriching the information recommended to the user. Moreover, by selecting the label in the obtained label pool according to the preset method, the label recommended to the user changes in real time, and the novelty of the recommendation information is enhanced.
  • the confidence interval upper bound algorithm selects a preset number of labels from the label pool to recommend to the user, so as to ensure that all types of labels are fully displayed, which can effectively prevent the Matthew effect, and at the same time ensure optimal label combination. Rendering, maximizing the usage of labels and the clickthrough rate of search listing pages.
  • a device for acquiring recommendation information disclosed in this embodiment is as shown in FIG. 3, and the device includes:
  • the prompt information obtaining module 300 is configured to obtain prompt information to be displayed
  • a tag pool construction module 310 configured to construct a tag pool based on the prompt information and a preset tag knowledge map
  • the recommendation module 320 is configured to select, from the tag pool constructed by the tag pool construction module 310, a preset number of tags to be recommended to the user;
  • the tag knowledge map is: a knowledge map in which the prompt information is a core word and a corresponding dimension information of the prompt information is described by a label.
  • the recommendation module 320 is further configured to:
  • a predetermined number of tags are selected from the tag pool by the confidence interval upper bound algorithm to be recommended to the user.
  • the recommendation module 320 further includes:
  • the expected benefit determining unit 3201 is configured to estimate a expected benefit of each tag in the tag pool according to historical behavior data of the tag by the user;
  • the adjustment indicator determining unit 3202 is configured to determine a revenue adjustment indicator of each label in the label pool according to the total number of times the history label is displayed and the total number of times each label in the label pool is displayed;
  • a recommended value determining unit 3203 configured to use a sum of the expected revenue and the income adjustment indicator as a recommended value of each label in the label pool;
  • the label selection unit 3204 is configured to select, from the label pool, a preset number of labels with the highest recommended value to the user.
  • the expected revenue determining unit 3201 includes any one of the following:
  • a first expected revenue determining subunit (not shown) for estimating a expected revenue of a corresponding label in the label pool according to a historical click rate of the same prompt information as the label;
  • the second expected revenue determining subunit (not shown) is configured to estimate a expected revenue of the corresponding label in the label pool according to the current user's user characteristics and the estimated click rate of the label.
  • the tag pool construction module 310 includes:
  • the core word matching unit 3101 is configured to match the prompt information with each core word in the preset label knowledge map
  • the tag pool establishing unit 3102 is configured to add all tags in the tag knowledge map corresponding to the successfully matched core words to the tag pool.
  • the apparatus further includes: a first knowledge map construction module 330, and/or a second knowledge map construction module 340;
  • the first knowledge map construction module 330 is configured to construct a label knowledge map according to the structured interest point data.
  • the first knowledge map construction module 330 includes:
  • a first core word and label determining unit 3301 configured to determine a core word and a label of the core word based on the point of interest name, the category name, and the additional attribute name in the structured point of interest data;
  • the first map establishing unit 3302 is configured to establish a tree relationship of each label according to the category system level relationship of the structured point of interest data, the additional attribute system level relationship, and the correspondence between the category system and the additional attribute system;
  • the root node of each of the tree relationships is a core word of the cluster tag knowledge map, and the leaf nodes are labels of corresponding levels.
  • the second knowledge map construction module 340 is configured to construct a label knowledge map according to the user behavior log.
  • the second knowledge map construction module 340 includes:
  • a second core word and label determining unit 3401 configured to determine a core word and a label of the core word based on a frequent item set of the mined user behavior log;
  • the second map establishing unit 3402 is configured to establish a tree relationship of the corresponding label according to the preset association relationship between the frequent items;
  • the root node of the tree relationship is a core word of the cluster tag knowledge map, and the leaf node is a tag;
  • the preset association relationship includes: an association relationship between the commodity and the merchant, and an association relationship between the merchant and the business circle.
  • the device for acquiring recommendation information disclosed in the embodiment of the present application by constructing a tag knowledge map in advance, and after acquiring the prompt information to be displayed, constructing a tag pool based on the prompt information and a preset tag knowledge map, and then A preset number of tags are selected in the tag pool and recommended to the user, enriching the information recommended to the user. Moreover, by selecting the label in the obtained label pool according to the preset method, the label recommended to the user changes in real time, and the novelty of the recommendation information is enhanced.
  • the confidence interval upper bound algorithm is used to select a preset number of labels from the label pool to recommend to the user, to ensure that all types of labels are fully displayed, to prevent the Matthew effect, and at the same time, to ensure the optimal label combination is presented. Maximize label usage and clickthrough rate of search listing pages.
  • the present application also discloses an electronic device including a memory, a processor, and a computer program stored on the memory and operable on the processor, the processor executing the computer program to implement the present application
  • the method for obtaining recommendation information according to the first embodiment and the second embodiment can be a PC, a mobile terminal, a personal digital assistant, a tablet, or the like.
  • the present application also discloses a computer readable storage medium, on which a computer program is stored, and when the program is executed by the processor, the steps of obtaining the recommendation information according to the first embodiment and the second embodiment of the present application are implemented.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Business, Economics & Management (AREA)
  • Computational Linguistics (AREA)
  • Finance (AREA)
  • Software Systems (AREA)
  • Accounting & Taxation (AREA)
  • Mathematical Physics (AREA)
  • Economics (AREA)
  • Artificial Intelligence (AREA)
  • General Business, Economics & Management (AREA)
  • Evolutionary Computation (AREA)
  • Computing Systems (AREA)
  • Technology Law (AREA)
  • Strategic Management (AREA)
  • Marketing (AREA)
  • Development Economics (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

本申请提供了一种推荐信息的获取方法，所述方法包括：获取待展示的提示信息；基于所述提示信息和预设的标签知识图谱构建标签池；从所述标签池中选择预设数量的标签推荐给用户。其中，所述标签知识图谱为基于核心词、以标签描述所述核心词的多维度信息的知识图谱。

Description

一种推荐信息的获取方法及装置、电子设备
相关申请的交叉引用
本专利申请要求于2017年7月26日提交的、申请号为201710618792.2、发明名称为“一种推荐信息的获取方法及装置，电子设备”的中国专利申请的优先权，该申请的全文以引用的方式并入本文中。
技术领域
本申请涉及计算机技术领域，特别是涉及一种推荐信息的获取方法及装置、电子设备。
背景技术
下拉提示(suggest)旨在对用户(在搜索框的)输入文本进行联想，推荐展示相关匹配提示词供用户选择，以降低用户输入成本，同时对联想结果进行质量把控，确保便于搜索引擎的理解和检索，从而提升用户搜索体验。下拉提示通常基于用户输入的文本进行前缀匹配，例如：根据输入前缀的权重及各权重下的各下拉提示结果的权重计算各推荐信息的推荐概率，如搜“火”suggest会给出“火锅”等推荐信息作为提示信息供用户选择。进一步的，为了扩充提供给用户的推荐信息扩充，还可针对每一个提示信息，推荐与该提示信息相关的类目信息供用户选择。例如，当用户输入搜索关键词“衬衫”时，提示信息中包括“衬衫连衣裙”，同时，系统根据预设的标签词库，选择提示信息“衬衫连衣裙”的类目标签，然后推荐提示信息的类目信息，如：长款、修身、韩版等，推荐信息较单一。
发明内容
本申请提供一种推荐信息的获取方法，以尽量丰富所获取的推荐信息。
为了解决上述问题，第一方面，本申请实施例提供了一种推荐信息的获取方法包括：
获取待展示的提示信息；
基于所述提示信息和预设的标签知识图谱构建标签池，其中，所述标签知识图 谱是基于核心词并以标签描述所述核心词的多维度信息的知识图谱；
从所述标签池中选择预设数量的标签推荐给用户。
第二方面，本申请实施例提供了一种推荐信息的获取装置，包括：
提示信息获取模块，用于获取待展示的提示信息；
标签池构建模块，用于基于所述提示信息和预设的标签知识图谱构建标签池，其中，所述标签知识图谱是基于核心词并以标签描述所述核心词的多维度信息的知识图谱；
推荐模块，用于从所述标签池中选择预设数量的标签推荐给用户。
第三方面，本申请实施例还公开了一种电子设备，包括存储器、处理器及存储在所述存储器上并可在所述处理器上运行的计算机程序，所述处理器执行所述计算机程序时实现本申请实施例所述的推荐信息的获取方法。
第四方面，本申请实施例提供了一种计算机可读存储介质，其上存储有计算机程序，该计算机程序被处理器执行时实现本申请实施例公开的推荐信息的获取方法的步骤。
本申请实施例公开的推荐信息的获取方法，通过获取待展示的提示信息，并基于所述提示信息和预设的标签知识图谱构建标签池，然后从所述标签池中选择预设数量的标签推荐给用户，丰富了推荐给用户的信息。并且，通过根据预设方法在获得的标签池中对标签进行选择，使得推荐给用户的标签实时变化，增强了推荐信息的新颖性。
附图说明
为了更清楚地说明本申请实施例的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本申请的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动性的前提下，还可以根据这些附图获得其他的附图。
图1是本申请一实施例的推荐信息的获取方法流程图。
图2是本申请另一实施例的推荐信息的获取方法流程图。
图3是本申请一实施例的推荐信息的获取装置结构的示意图。
图4是本申请另一实施例的推荐信息的获取装置结构的示意图。
图5是本申请又一实施例的推荐信息的获取装置结构的示意图。
具体实施方式
下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本申请的一部分实施例，而不是全部的实施例。基于本申请中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本申请保护的范围。
本申请实施例所述的推荐信息的获取方法可以应用于搜索推荐领域。例如，用户通过应用页面输入查询词之后，搜索引擎基于前缀搜索，召回多个搜索结果，并展示为提示信息。对于每个提示信息，推荐系统通过调用本申请公开的推荐信息获取方法获取相应的推荐信息，并在适当的位置进行展示。
实施例一
本实施例公开的一种推荐信息的获取方法，如图1所示，该方法包括：步骤100至步骤120。
步骤100，获取待展示的提示信息。
在推荐应用中，当用户输入搜索关键词之后，搜索引擎会给予前缀匹配的方式，返回相应的搜索结果，用于展示给用户。然后，推荐应用按照搜索结果的匹配度或点击率等指标，选择指标最好的预设数量的搜索结果作为提示信息，便于推荐应用在搜索页面展示给用户。具体实施时，检测到用户输入搜索关键词之后，通过调用搜索引擎提供的接口可以获得待展示给用户的提示信息。
步骤110，基于所述提示信息和预设的标签知识图谱构建标签池。
其中，所述标签知识图谱为：以所述提示信息为核心词、以标签描述所述提示信息的相应维度信息的知识图谱。
在基于提示信息进行信息推荐之前，首先要建立标签知识图谱。
建立标签知识图谱时，首先要确定可能的提示信息。提示信息通常基于用户的搜索日志和基于搜索结果的点击日志获得。例如，很多用户在输入“汽车”之后，点击了“汽车美容”这个搜索结果，则建立标签知识图谱时，“汽车美容”将作为一个提示信息。
标签知识图谱是以提示信息为核心词、以标签描述所述提示信息的相应维度信息的知识图谱。标签知识图谱中包括多个核心词，围绕每一个核心词，构建了该核心词的多维度的知识体系，最后形成簇状的知识图谱。其中，标签用于描述该核心词的各个维度的信息，例如，对于核心词“汽车美容”，其类目维度的信息可以描述为“内部美容”、“漆面护理”，其品牌维度的信息可以秒速描述为“汽车工坊”、“汽车亿佰”，则提示信息“汽车美容”的标签包括：“内部美容”、“漆面护理”、“汽车工坊”、“汽车亿佰”。然后，根据通过数据结构表示提示信息和各标签的层级及类目隶属关系，构成每个所述提示信息的知识图谱。具体实施时，标签涉及的维度包括：类目、子类目、附加属性等，附加属性可以为：商家信息、品牌信息、地域信息、产品特征信息等。核心词以及核心词的标签通常通过对平台数据的分析，结合具体业务需求确定。
具体实施时，标签知识图谱可以采用树状结构来表示。树状结构的根节点是每个簇状知识图谱的核心词，树状结构的叶子节点是核心词的各种维度的标签。通过将所述提示信息和预设的标签知识图谱中的每个核心词进行匹配，然后，取匹配成功的核心词对应的标签知识图谱中的所有标签作为该提示信息的标签，加入标签池。
步骤120，从所述标签池中选择预设数量的标签推荐给用户。
鉴于展示空间的局限，通常从所述标签池中选择预设数量的标签推荐给用户，如每个提示信息展示3个推荐标签。在选择展示给用户的标签时，可以随机选出预设数量的标签展示给用户；还可以挑选点击率高的标签展示给用户；或者，通过置信区间上界算法从所述标签池中选择预设数量的推荐值最高的标签推荐给用户。在计算推荐值的过程中可以根据标签的历史点击数据，或者结合标签的历史点击数据和接受推荐的用户的用户特征计算标签的推荐值。
本申请实施例公开的推荐信息的获取方法，通过获取待展示的提示信息，并基于所述提示信息和预设的标签知识图谱构建标签池，然后从所述标签池中选择预设数量的标签推荐给用户，解决了现有技术中存在的获取的推荐信息单一，且相对于预设标签词库固定不变的问题。通过基于预设的标签知识图谱构建标签池，丰富了推荐给用户的信息。通过根据预设方法在获得的标签池中对标签进行选择，使得推荐给用户的标签实时变化，增强了推荐信息的新颖性。
实施例二
本实施例公开的一种推荐信息的获取方法，如图2所示，该方法包括：步骤200 至步骤230。
步骤200，构建标签知识图谱。
在基于提示信息进行信息推荐之前，首先要建立标签知识图谱。
建立标签知识图谱时，首先要确定可能的提示信息。提示信息通常基于用户的搜索日志和搜索结果的点击日志获得。例如，很多用户在输入“汽车”之后，点击了“汽车美容”这个搜索结果，则建立标签知识图谱时，“汽车美容”将作为一个提示信息。
标签知识图谱是以提示信息为核心词、以标签描述所述提示信息的相应维度信息的知识图谱。其中标签是与核心词相关的词语，如描述核心词的各个维度属性的词语。标签知识图谱中包括多个核心词、围绕每一个核心词的标签，核心词和标签之间的层级关联关系构成了该核心词的多维度的知识体系，最后形成簇状的知识图谱。其中，标签用于描述该核心词的各个维度的信息，例如，对于核心词“汽车美容”，其类目维度的信息可以描述为“内部美容”、“漆面护理”，其品牌维度的信息可以描述为“汽车工坊”、“汽车亿佰”，则提示信息“汽车美容”的标签包括：“内部美容”、“漆面护理”、“汽车工坊”、“汽车亿佰”。然后，根据通过数据结构表示提示信息和各标签的层级及类目隶属关系，构成每个所述提示信息的知识图谱。具体实施时，标签涉及的维度包括：类目、子类目、附加属性等，附加属性可以为：商家信息、品牌信息、地域信息、产品特征信息等。核心词以及核心词的标签通常通过对平台数据的分析，结合具体业务需求确定。
构建标签知识图谱包括：根据结构化兴趣点数据构建标签知识图谱，和/或，根据用户行为日志构建标签知识图谱。具体实施时，可以仅根据结构化兴趣点数据构建标签知识图谱，也可以仅根据用户行为日志构建标签知识图谱，还可以根据结构化兴趣点数据和用户行为日志这两者构建标签知识图谱。
当根据结构化兴趣点数据构建标签知识图谱时，构建标签知识图谱的具体方法，包括：基于结构化兴趣点数据中的兴趣点名称、类目名称和附加属性名称，确定核心词和所述核心词的标签；根据所述结构化兴趣点数据的类目体系层级关系、附加属性体系层级关系，以及类目体系和附加属性体系的对应关系，建立各标签的树状关系；其中，每个所述树状关系的根节点为该簇标签知识图谱的核心词，叶子节点为相应层级的标签。
结构化兴趣点(Point of Interest)数据，以下简称POI数据，通常是搜索或推荐平台描述商品信息的数据，POI数据通常包括：类目体系和附加属性体系。
类目体系是按照业务需求，对商品设置不同的类目标签，以及所述类目的层级关系。类目标签包括：父类目类目名称、子类目类目名称；类目层级关系包括：所属类目、所属类目的父类目、所属类目包含的子类目等。以“丽人”业务为例，其类目体系包括的父类目有：美甲、美发、瑜伽、舞蹈、纹身等；美发类目下面又进一步划分为：染发、理发、发型设计等多个子类目，依次往下，构成了层级类目关系。基于POI数据的类目体系可以获取一定数量的商品的知识标签图谱。
附加属性体系是按照业务需求，对商品设置不同的附加属性标签，以及所述附加属性的层级关系。以“美甲”产品为例，其中，“美甲”为兴趣点名称，其附加属性可以包括:彩绘、磨砂等附加属性标签。“彩绘”标签的下一个级别又包括：富贵、清新、田园等属性。类目体系与附加属性体系的信息互相补充，可以得到构建标签知识谱图的更全面的数据。
每条POI数据包括多个字段，分别表示POI名称、所属父类目、所属子类目、属性信息等。具体实施时，通过对POI数据进行解析，可以获取POI名称对应的类目名称、各类目名称的层级关系、附加属性名称、各属性名称的层级关系等。其中，类目体系和附加属性体系中有具有关联关系，例如，“美甲”的类目层级为父类目，“彩绘”的类目层级为子类目。基于从POI数据中提取的上述信息可以构建标签知识图谱。
具体实施时，以核心词为“美甲”为例，其下级标签包括：“彩绘”和“磨砂”，据此，可以构建“美甲”的标签知识图谱。若以树状结构表示该标签知识图谱，其根节点为“美甲”，叶子节点包括：“彩绘”和“磨砂”。具体实施时，还可以进一步将描述富贵、清新、田园等附加属性标签作为叶子节点——“彩绘”的叶子节点，完善标签知识图谱。
当根据用户行为日志构建标签知识图谱时，构建标签知识图谱的具体方法，包括：基于挖掘的用户行为日志的频繁项集，确定核心词和所述核心词的标签；根据各频繁项之间的预设关联关系，建立相应标签的树状关系；其中，所述树状关系的根节点为该簇标签知识图谱的核心词，叶子节点为标签；所述预设关联关系包括：商品和商家的关联关系，商家和商圈的关联关系。
对于搜索平台或者推荐平台，平台上的应用拥有大量的用户行为日志，通过分析海量用户历史行为数据，可以获得核心词以及核心词相关的标签。具体实施时，对于用户行为日志，采用挖掘算法挖掘日志数据中的频繁项集，从挖掘到的频繁项集中选择核心词以及该核心词的标签，具体实施时，可以将搜索词作为核心词，根据具体业务需 求，从挖掘到的频繁项中选择该核心词的标签，并建立核心词到标签的层级关系。比如，搜索小龙虾的用户大量点击“沪小胖”，“沪小胖”就构成了小龙虾的标签。如果将“小龙虾”作为树状关系的根节点，将“沪小胖”作为树状关系的叶子节点，就构成了核心词“小龙虾”的标签知识图谱的一部分。
具体实施时，还可以通过对用户行为日志进行频繁项集挖掘，挖掘到大部分用户搜索某个词时点击的商家后，进一步计算点击的商家的分布，然后，将商家作为核心词，将商家分布的商圈作为商家的标签，建立由商家到商圈的层级关系，进一步构建标签知识图谱。即将商家作为树状关系的根节点，将商圈作为树状关系的叶子节点。
具体实施时，通过频繁项集挖掘方法，可以根据业务需求选择核心词和标签，并进一步建立标签核心词的标签知识图谱。预设的关联关系还可以根据具体业务需求设置为其他关系，此处不一一列举。
通过基于用户行为日志进行频繁项集挖掘，从中挖掘出品质高、点击率高的商户以及优质商圈推荐给用户，进一步提升了用户体验。
步骤210，获取待展示的提示信息。
在推荐应用中，当用户输入搜索关键词之后，搜索引擎会给予前缀匹配的方式，返回相应的搜索结果，用于展示给用户。然后，推荐应用按照搜索结果的匹配度或点击率等指标，选择指标最好的预设数量的搜索结果作为提示信息，便于推荐应用在搜索页面上展示给用户。具体实施时，检测到用户输入搜索关键词之后，通过调用搜索引擎提供的接口可以获得待展示给用户的提示信息。具体实施时，所述提示信息可以采用下拉提示词的形式展示给用户，也可以采用其他形式展示给用户。
步骤220，基于所述提示信息和预设的标签知识图谱构建标签池。
其中，所述标签知识图谱为：以所述提示信息为核心词、以标签描述所述提示信息的相应维度信息的知识图谱。
具体实施时，所述基于所述提示信息和预设的标签知识图谱构建标签池，包括：将所述提示信息和预设的标签知识图谱中的每个核心词进行匹配；将匹配成功的核心词对应的标签知识图谱中的所有标签加入标签池。
具体实施时，以标签知识图谱采用树状结构来表示为例。树状结构的根节点是每个簇状知识图谱的核心词，树状结构的叶子节点是核心词的各种维度的标签。通过将所述提示信息和预设的标签知识图谱中的每个根节点的核心词进行匹配，然后，取匹配 成功的根节点下所有叶子节点对应的标签作为该提示信息的标签，加入标签池。
每一个提示信息对应的标签数量与该提示信息所在知识图谱的层级有关，层级越高，其对应的标签越多。具体实施时，受显示空间的限制，针对每个提示信息，显示给用户的推荐词，即标签的数量是有限的。例如：限定每个下拉词最多能展示三个与之相关标签，同时限制整个屏幕，最多允许展示九个标签，以避免屏幕标签泛滥。因此，通常按照知识图谱层次由高到低的顺序，取所述提示信息匹配成功的核心词所在知识图谱的最高几层中的标签构建标签池。
步骤230，从所述标签池中选择预设数量的标签推荐给用户。
鉴于展示空间的局限，通常从所述标签池中选择预设数量的标签推荐给用户，如每个提示信息展示3个推荐标签。在选择展示给用户的标签时，可以随机选出预设数量的标签展示给用户，但是随即选择的方法会导致许多冷门标签曝光过多，浪费了宝贵的展示机会；还可以挑选点击率高的标签展示给用户，这种方法带来的缺陷是展示给每个用户的标签是相同的，同时大量的标签没有展示机会，损害了用户体验。
优选的，具体实施时，所述从所述标签池中选择预设数量的标签推荐给用户，包括：通过置信区间上界算法从所述标签池中选择预设数量的标签推荐给用户。
所述通过置信区间上界算法从所述标签池中选择预设数量的标签推荐给用户，进一步包括：根据用户对标签的历史行为数据预估所述标签池中每个标签的期望收益；根据所述提示信息的历史标签展示总次数和所述标签池中每个标签的展示总次数，分别确定所述标签池中每个标签的收益调整指标；将所述期望收益和所述收益调整指标的和作为所述标签池中各标签的推荐值；从所述标签池中选择推荐值最高的预设数量标签推荐给用户。
具体实施时，标签的期望收益revenue可以表示为：
revenue＝X j(t)，       公式1
其中，t为提示信息X的标签j的期望收益。
所述根据用户对标签的历史行为数据预估所述标签池中每个标签的期望收益通过两种方法来实现。
第一种，根据与标签相同的提示信息的历史点击率，预估所述标签池中相应标签的期望收益。
以推荐应用的搜索页面为例，在展示根据搜索关键词搜索得到的推荐信息时，通常设置提示信息展示位置和标签展示位置。其中，提示信息展示位置用于展示根据搜索关键词搜索得到的提示信息，如下拉提示词；标签展示位置用于展示提示信息的标签，如下拉提示词的类别。当根据提示信息确定的标签没有在标签展示位置展示过时，可以根据该标签作为提示信息时被展示后的历史点击率，预估所述标签池中相应标签的期望收益。计算提示信息的历史点击率的具体方法参见本领域技术人员熟知的任意技术，例如，参见计算下拉提示词的历史点击率的具体方法，此处不再赘述。
以用户输入“汽车”之后，搜索引擎返回的提示信息包括：“汽车”、“汽车美容”为例。如果推荐应用针对提示信息“汽车美容”确定的标签包括：“汽车亿佰”、“汽车工坊”、“汽车驿站”等标签，假设其中的“汽车亿佰”之前没有作为提示信息的标签被推荐给用户，则根据搜索引擎的历史数据，确定提示信息“汽车亿佰”的历史点击率，然后，根据提示信息“汽车亿佰”的历史点击率确定标签“汽车亿佰”的期望收益。例如，将提示信息“汽车亿佰”的历史点击率作为标签“汽车亿佰”的期望收益，或者，将提示信息“汽车亿佰”的历史点击率乘以一个系数，然后将乘积作为标签“汽车亿佰”的期望收益。具体实施时，根据与标签相同的提示信息的历史点击率，预估所述标签池中相应标签的期望收益revenue时，例如公式1中的t可为提示信息“汽车美容”的标签“汽车亿佰”的期望收益，即“汽车亿佰”作为提示信息时的历史点击率。还可以根据具体业务需求，采用其他计算方法，本申请对具体的计算方法不做限定。
第二种，根据当前用户的用户特征和标签的预估点击率，预估所述标签池中相应标签的期望收益。
优选的，为了进一步提升用户体验，对不同的用户推荐适合的标签，具体实施时，可以根据当前用户的用户特征和标签的预估点击率，预估所述标签池中相应标签的期望收益。当标签被展示后，标签的历史点击率数据将会实时更新，根据更新后的历史点击率数据进一步计算标签的期望收益，将会提高计算结果的准确性。具体实施时，当收集到大量的用户对标签是否点击的行为数据后，可以用逻辑回归算法(LR，Logistic Regression)训练出用户点击概率模型，更准确的计算出用户的当前的期望收益。训练点击概率模型的具体方法参见本领域技术人员熟知的任意技术，此处不再赘述。
然后，可根据如下公式2预估标签的期望收益：
Figure PCTCN2017120045-appb-000001
其中，φ是用户u的特征和提示信息X的标签j的预估点击率特征构成的特征向量。其中，用户特征包括但不限于：用户搜索地理位置、搜索时段、用户对标签的类目偏好、用户对推荐商户的偏好等中的一项或多项。预估点击率特征可以为标签j的历史点击次数或根据用户行为数据预估的点击率。W是逻辑回归算法LR的参数向量，与用户的特征向量相对应，W具体为特征向量φ中每个维度特征的权重。T为与φ和W对应的转置矩阵。
具体实施时，可根据用户u对标签的历史点击数据，即对标签j是否点击的行为数据，使用开源LR算法包来训练得到模型W向量。然后，结合用户u的特征向量φ估计用户的期望收益，实现有针对性的对用户推荐标签，并且随着点击行为的变化，提供多样化的推荐标签。用户的特征向量φ的具体提取方法，可以参见本领域技术人员熟知的任意技术，此处不再赘述。
本申请实施例中，用户的特征向量至少包括点击率特征。每个标签的收益调整指标bonus用来根据业务的需求，平衡标签的展示几率。具体实施时，根据所述提示信息的历史标签展示总次数和所述标签池中每个标签的展示总次数，分别确定所述标签池中每个标签的收益调整指标。在本申请的一个实施例中，可以采用如下公式3计算收益调整指标bonus。
Figure PCTCN2017120045-appb-000002
上述公式中，t是针对该提示信息X展示的标签的历史总次数；T j,t是在展示提示信息X的标签的过程中，标签j的展示次数。
由上述公式3可以看出，对于展示次数越少的标签，也会得到展示机会，并且会适当提高推荐值；而展示次数越多的标签，会被适当调低推荐值，给展示次数少的标签分配一些展示机会。具体实施时，还可以采用其他表示的t和T j,t的反比例函数计算收益调整指标，此处不一一列举。
然后，将所述期望收益和所述收益调整指标之和作为所述标签池中各标签的推荐值，即推荐值＝revenue+bonus。在确定标签的展示得分的过程中，标签的期望收益值占较大权重，而标签的收益调整指标bonus仅起到微调的作用。因此，具体实施时，可以通过设置加权求和的权重系数或者调整收益调整指标的计算公式来平衡期望收益和 收益调整指标在计算推荐值时的贡献。
最后，从所述标签池中选择推荐值最高的预设数量标签推荐给用户。以预设数量等于2为例，假设提示信息“汽车美容”确定的标签包括：“汽车亿佰”、“汽车工坊”、“汽车驿站”等标签，经过计算，选择推荐值最高的标签“汽车亿佰”、“汽车工坊”推荐给用户。
本申请实施例公开的推荐信息的获取方法，通过预先构建标签知识图谱，并在获取到待展示的提示信息后，基于所述提示信息和预设的标签知识图谱构建标签池，然后从所述标签池中选择预设数量的标签推荐给用户，丰富了推荐给用户的信息。并且，通过根据预设方法在获得的标签池中对标签进行选择，使得推荐给用户的标签实时变化，增强了推荐信息的新颖性。
进一步的，通过置信区间上界算法从所述标签池中选择预设数量的标签推荐给用户，最大程度保证各类标签得到充分展示，可有效防止马太效应，同时，确保最优标签组合的呈现，最大化提升标签的使用率和搜索列表页的点击率。
实施例三
本实施例公开的一种推荐信息的获取装置，如图3所示，所述装置包括：
提示信息获取模块300，用于获取待展示的提示信息；
标签池构建模块310，用于基于所述提示信息和预设的标签知识图谱构建标签池；
推荐模块320，用于从所述标签池构建模块310构建的标签池中选择预设数量的标签推荐给用户；
其中，所述标签知识图谱为：以所述提示信息为核心词、以标签描述所述提示信息相应维度信息的知识图谱。
可选的，所述推荐模块320进一步用于：
通过置信区间上界算法从所述标签池中选择预设数量的标签推荐给用户。
可选的，如图4所示，所述推荐模块320进一步包括：
期望收益确定单元3201，用于根据用户对标签的历史行为数据预估所述标签池中每个标签的期望收益；
调整指标确定单元3202，用于根据所述提示信息的历史标签展示总次数和所述 标签池中每个标签的展示总次数，分别确定所述标签池中每个标签的收益调整指标；
推荐值确定单元3203，用于将所述期望收益和所述收益调整指标之和作为所述标签池中各标签的推荐值；
标签选择单元3204，用于从所述标签池中选择推荐值最高的预设数量标签推荐给用户。
可选的，所述期望收益确定单元3201包括以下任意一项：
第一期望收益确定子单元(图中未示出)，用于根据与标签相同的提示信息的历史点击率，预估所述标签池中相应标签的期望收益；
第二期望收益确定子单元(图中未示出)，用于根据当前用户的用户特征和标签的预估点击率，预估所述标签池中相应标签的期望收益。
可选的，如图4所示，所述标签池构建模块310包括：
核心词匹配单元3101，用于将所述提示信息和预设的标签知识图谱中的每个核心词进行匹配；
标签池建立单元3102，用于将匹配成功的核心词对应的标签知识图谱中的所有标签加入标签池。
可选的，如图5所示，所述装置还包括：第一知识图谱构建模块330，和/或，第二知识图谱构建模块340；其中，
所述第一知识图谱构建模块330，用于根据结构化兴趣点数据构建标签知识图谱。
可选的，如图5所示，所述第一知识图谱构建模块330包括：
第一核心词和标签确定单元3301，用于基于结构化兴趣点数据中的兴趣点名称、类目名称和附加属性名称，确定核心词和所述核心词的标签；
第一图谱建立单元3302，用于根据所述结构化兴趣点数据的类目体系层级关系、附加属性体系层级关系，以及类目体系和附加属性体系的对应关系，建立各标签的树状关系；其中，每个所述树状关系的根节点为该簇标签知识图谱的核心词，叶子节点为相应层级的标签。
所述第二知识图谱构建模块340，用于根据用户行为日志构建标签知识图谱。
可选的，如图5所示，所述第二知识图谱构建模块340包括：
第二核心词和标签确定单元3401，用于基于挖掘的用户行为日志的频繁项集，确定核心词和所述核心词的标签；
第二图谱建立单元3402，用于根据各频繁项之间的预设关联关系，建立相应标签的树状关系；
其中，所述树状关系的根节点为该簇标签知识图谱的核心词，叶子节点为标签；所述预设关联关系包括：商品和商家的关联关系，商家和商圈的关联关系。
本申请实施例公开的推荐信息的获取装置，通过预先构建标签知识图谱，并在获取到待展示的提示信息后，基于所述提示信息和预设的标签知识图谱构建标签池，然后从所述标签池中选择预设数量的标签推荐给用户，丰富了推荐给用户的信息。并且，通过根据预设方法在获得的标签池中对标签进行选择，使得推荐给用户的标签实时变化，增强了推荐信息的新颖性。
进一步的，通过置信区间上界算法从所述标签池中选择预设数量的标签推荐给用户，最大程度保证各类标签得到充分展示，防止马太效应，同时，确保最优标签组合的呈现，最大化提升标签的使用率和搜索列表页的点击率。
相应的，本申请还公开了一种电子设备，包括存储器、处理器及存储在所述存储器上并可在处理器上运行的计算机程序，所述处理器执行所述计算机程序时实现如本申请实施例一和实施例二所述的推荐信息的获取方法。所述电子设备可以为PC机、移动终端、个人数字助理、平板电脑等。
本申请还公开了一种计算机可读存储介质，其上存储有计算机程序，该程序被处理器执行时实现如本申请实施例一和实施例二所述的推荐信息的获取方法的步骤。
本说明书中的各个实施例均采用递进的方式描述，每个实施例重点说明的都是与其他实施例的不同之处，各个实施例之间相同或相似的部分互相参见即可。对于装置实施例而言，由于其与方法实施例基本相似，所以描述的比较简单，相关之处参见方法实施例的部分说明即可。
以上对本申请提供的一种推荐信息的获取方法及装置进行了详细介绍，本文中应用了具体实例对本申请的原理及实施方式进行了阐述，以上实施例的说明只是用于帮助理解本申请的方法及其核心思想；同时，对于本领域的一般技术人员，依据本申请的思想，在具体实施方式及应用范围上均会有改变之处，综上所述，本说明书内容不应理解为对本申请的限制。
通过以上的实施方式的描述，本领域的技术人员可以清楚地了解到各实施方式可借助软件和必需的通用硬件平台的方式来实现，当然也可以通过硬件实现。基于这样的理解，上述技术方案本质上或者说对现有技术做出贡献的部分可以以软件产品的形式体现出来，该计算机软件产品可以存储在计算机可读存储介质中，如ROM/RAM、磁碟、光盘等，包括若干指令用以使得一台计算机设备(可以是个人计算机，服务器，或者网络设备等)执行各个实施例或者实施例的某些部分所述的方法。

Claims (18)

  1. 一种推荐信息的获取方法，包括：
    获取待展示的提示信息；
    基于所述提示信息和预设的标签知识图谱构建标签池，其中，所述标签知识图谱是基于核心词并以标签描述所述核心词的多维度信息的知识图谱；
    从所述标签池中选择预设数量的标签推荐给用户。
  2. 根据权利要求1所述的方法，其特征在于，从所述标签池中选择预设数量的标签推荐给用户，包括：
    通过置信区间上界算法从所述标签池中选择预设数量的标签推荐给用户。
  3. 根据权利要求2所述的方法，其特征在于，通过所述置信区间上界算法从所述标签池中选择预设数量的标签推荐给用户，包括：
    根据用户对所述标签池中每个标签的历史行为数据，预估所述标签池中每个标签的期望收益；
    根据所述提示信息的历史标签展示总次数和所述标签池中每个标签的展示总次数，确定所述标签池中每个标签的收益调整指标；
    将所述期望收益和所述收益调整指标的加和，作为所述标签池中各标签的推荐值；
    从所述标签池中选择所述推荐值最高的预设数量的标签推荐给用户。
  4. 根据权利要求3所述的方法，其特征在于，用户对所述标签的历史行为数据包括以下任意一个或多个：
    与所述标签相同的提示信息的历史点击率；
    当前用户的用户特征；
    所述标签的预估点击率。
  5. 根据权利要求1所述的方法，其特征在于，基于所述提示信息和所述标签知识图谱构建所述标签池，包括：
    将所述提示信息和所述标签知识图谱中的每个核心词进行匹配；
    将匹配成功的核心词对应的所有标签加入所述标签池。
  6. 根据权利要求1至5任一项所述的方法，其特征在于，还包括以下任意一个或 多个：
    根据结构化兴趣点数据构建所述标签知识图谱，
    根据用户行为日志构建所述标签知识图谱。
  7. 根据权利要求6所述的方法，其特征在于，根据所述结构化兴趣点数据构建所述标签知识图谱，包括：
    基于所述结构化兴趣点数据中的兴趣点名称、类目名称和附加属性名称，确定核心词和所述核心词的标签；
    根据所述结构化兴趣点数据的类目体系的层级关系、附加属性体系的层级关系，以及所述类目体系和所述附加属性体系之间的对应关系，建立各所述标签的树状关系；
    其中，每个所述树状关系的根节点为该簇标签知识图谱的核心词，叶子节点为相应层级的标签。
  8. 根据权利要求6所述的方法，其特征在于，根据所述用户行为日志构建所述标签知识图谱，包括：
    基于挖掘的用户行为日志的频繁项集，确定核心词和所述核心词的标签；
    根据各频繁项之间的预设关联关系，建立标签的树状关系，其中，所述预设关联关系包括商品和商家的关联关系、商家和商圈的关联关系；
    其中，所述树状关系的根节点为该簇标签知识图谱的核心词，叶子节点为标签。
  9. 一种推荐信息的获取装置，包括：
    提示信息获取模块，用于获取待展示的提示信息；
    标签池构建模块，用于基于所述提示信息和预设的标签知识图谱构建标签池，其中，所述标签知识图谱是基于核心词并以标签描述所述核心词的多维度信息的知识图谱；
    推荐模块，用于从所述标签池中选择预设数量的标签推荐给用户。
  10. 根据权利要求9所述的装置，其特征在于，所述推荐模块进一步用于：
    通过置信区间上界算法从所述标签池中选择预设数量的标签推荐给用户。
  11. 根据权利要求10所述的装置，其特征在于，所述推荐模块进一步包括：
    期望收益确定单元，用于根据用户对所述标签池中每个标签的历史行为数据，预估所述标签池中每个标签的期望收益；
    调整指标确定单元，用于根据所述提示信息的历史标签展示总次数和所述标签池中每个标签的展示总次数，确定所述标签池中每个标签的收益调整指标；
    推荐值确定单元，用于将所述期望收益和所述收益调整指标的加和，作为所述标签池中各标签的推荐值；
    标签选择单元，用于从所述标签池中选择推荐值最高的预设数量的标签推荐给用户。
  12. 根据权利要求11所述的装置，其特征在于，用户对所述标签的历史行为数据包括以下任意一个或多个：
    与所述标签相同的提示信息的历史点击率；
    当前用户的用户特征；和
    所述标签的预估点击率。
  13. 根据权利要求9所述的装置，其特征在于，所述标签池构建模块包括：
    核心词匹配单元，用于将所述提示信息和所述标签知识图谱中的每个核心词进行匹配；
    标签池建立单元，用于将匹配成功的核心词对应的所有标签加入所述标签池。
  14. 根据权利要求9至13任一项所述的装置，其特征在于，所述装置还包括以下任意一个或多个：
    第一知识图谱构建模块，用于根据结构化兴趣点数据构建所述标签知识图谱；
    第二知识图谱构建模块，用于根据用户行为日志构建所述标签知识图谱。
  15. 根据权利要求14所述的装置，其特征在于，所述第一知识图谱构建模块包括：
    第一核心词和标签确定单元，用于基于所述结构化兴趣点数据中的兴趣点名称、类目名称和附加属性名称，确定核心词和各所述核心词的标签；
    第一图谱建立单元，用于根据所述结构化兴趣点数据的类目体系的层级关系、附加属性体系的层级关系，以及所述类目体系和所述附加属性体系之间的对应关系，建立各所述标签的树状关系；
    其中，每个所述树状关系的根节点为该簇标签知识图谱的核心词，叶子节点为相应层级的标签。
  16. 根据权利要求14所述的装置，其特征在于，所述第二知识图谱构建模块包括：
    第二核心词和标签确定单元，用于基于挖掘的用户行为日志的频繁项集，确定核心词和各所述核心词的标签；
    第二图谱建立单元，用于根据各频繁项之间的预设关联关系，建立相应标签的树状关系，其中，所述预设关联关系包括商品和商家的关联关系、商家和商圈的关联关系；
    其中，所述树状关系的根节点为该簇标签知识图谱的核心词，叶子节点为标签。
  17. 一种电子设备，包括存储器、处理器及存储在所述存储器上并可在所述处理器上运行的计算机程序，其特征在于，所述处理器执行所述计算机程序时实现权利要求1至8任意一项所述的推荐信息的获取方法。
  18. 一种计算机可读存储介质，其上存储有计算机程序，其特征在于，该计算机程序被处理器执行时实现权利要求1至8中任意一项所述的推荐信息的获取方法的步骤。
PCT/CN2017/120045 2017-07-26 2017-12-29 一种推荐信息的获取方法及装置、电子设备 WO2019019554A1 (zh)

Priority Applications (5)

Application Number Priority Date Filing Date Title
CA3066941A CA3066941A1 (en) 2017-07-26 2017-12-29 Method and apparatus for obtaining recommendation information, and electronic device
EP17918888.3A EP3623956A4 (en) 2017-07-26 2017-12-29 PROCESS AND APPARATUS FOR OBTAINING RECOMMENDED INFORMATION, AND ELECTRONIC DEVICE
JP2019569754A JP7065122B2 (ja) 2017-07-26 2017-12-29 推奨情報の取得方法及び装置、電子機器
KR1020197036939A KR102370408B1 (ko) 2017-07-26 2017-12-29 추천 정보 획득 방법, 장치 및 전자설비
US16/715,085 US20200117675A1 (en) 2017-07-26 2019-12-16 Obtaining of Recommendation Information

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201710618792.2A CN107688606A (zh) 2017-07-26 2017-07-26 一种推荐信息的获取方法及装置，电子设备
CN201710618792.2 2017-07-26

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US16/715,085 Continuation US20200117675A1 (en) 2017-07-26 2019-12-16 Obtaining of Recommendation Information

Publications (1)

Publication Number Publication Date
WO2019019554A1 true WO2019019554A1 (zh) 2019-01-31

Family

ID=61153055

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/120045 WO2019019554A1 (zh) 2017-07-26 2017-12-29 一种推荐信息的获取方法及装置、电子设备

Country Status (7)

Country Link
US (1) US20200117675A1 (zh)
EP (1) EP3623956A4 (zh)
JP (1) JP7065122B2 (zh)
KR (1) KR102370408B1 (zh)
CN (1) CN107688606A (zh)
CA (1) CA3066941A1 (zh)
WO (1) WO2019019554A1 (zh)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110727859A (zh) * 2019-09-12 2020-01-24 北京十分科技有限公司 一种推荐信息推送方法及其装置
CN112288512A (zh) * 2020-10-09 2021-01-29 北京三快在线科技有限公司 信息处理方法、装置、电子设备及可读存储介质
CN112308432A (zh) * 2020-11-03 2021-02-02 重庆理工大学 一种基于类别和能力评价的众包任务推荐方法
CN113435197A (zh) * 2021-06-25 2021-09-24 平安国际智慧城市科技股份有限公司 数据推送方法、装置、推送服务器及存储介质
CN113626575A (zh) * 2021-09-01 2021-11-09 浙江力石科技股份有限公司 一种基于用户问答的智能推荐方法

Families Citing this family (42)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110309312B (zh) * 2018-03-09 2022-02-11 北京国双科技有限公司 一种关联事件获取方法及装置
CN108595526B (zh) * 2018-03-28 2020-10-23 北京奇艺世纪科技有限公司 资源推荐方法及装置
CN108717407B (zh) * 2018-05-11 2022-08-09 北京三快在线科技有限公司 实体向量确定方法及装置，信息检索方法及装置
CN108763556A (zh) * 2018-06-01 2018-11-06 北京奇虎科技有限公司 基于需求词的用户挖掘方法及装置
CN108763555A (zh) * 2018-06-01 2018-11-06 北京奇虎科技有限公司 基于需求词的画像数据获取方法及装置
CN108932530A (zh) * 2018-06-29 2018-12-04 新华三大数据技术有限公司 标签体系的构建方法及装置
CN110750561A (zh) * 2018-07-20 2020-02-04 深圳市诚壹科技有限公司 一种关联应用程序挖掘的方法及装置
CN109308332B (zh) * 2018-08-07 2022-05-20 腾讯科技（深圳）有限公司 一种目标用户获取方法、装置和服务器
CN110968776B (zh) * 2018-09-30 2023-05-23 北京国双科技有限公司 政策知识的推荐方法、装置存储介质及处理器
CN109447786A (zh) * 2018-10-24 2019-03-08 中国银行股份有限公司 信息录入方法及装置
CN109522345A (zh) * 2018-11-19 2019-03-26 百度在线网络技术（北京）有限公司 用于显示信息的方法及装置
CN111383042A (zh) * 2018-12-30 2020-07-07 贝壳技术有限公司 房源推荐方法和装置
CN111522886B (zh) * 2019-01-17 2023-05-09 中国移动通信有限公司研究院 一种信息推荐方法、终端及存储介质
CN109961357B (zh) * 2019-03-25 2021-09-03 上海拉扎斯信息科技有限公司 用户数据处理方法、装置、电子设备及存储介质
CN110188248A (zh) * 2019-05-28 2019-08-30 新华网股份有限公司 基于新闻问答交互系统的数据处理方法、装置及电子设备
CN110443675A (zh) * 2019-06-27 2019-11-12 北京三快在线科技有限公司 确定订单生命周期的方法、装置、电子设备及存储介质
CN110413881B (zh) * 2019-07-11 2023-10-20 腾讯科技（深圳）有限公司 一种识别标签准确性的方法、装置、网络设备和存储介质
CN110442700A (zh) * 2019-08-12 2019-11-12 珠海格力电器股份有限公司 用于人机交互的人机多轮对话方法及系统、智能设备
CN112395261A (zh) * 2019-08-16 2021-02-23 中国移动通信集团浙江有限公司 业务推荐方法、装置、计算设备及计算机存储介质
CN110766456A (zh) * 2019-10-16 2020-02-07 无线生活（杭州）信息科技有限公司 商品推荐方法及装置
CN110795555B (zh) * 2019-10-31 2022-06-17 北京明略软件系统有限公司 知识图谱数据展示方法及相关装置
CN113111251A (zh) * 2020-01-10 2021-07-13 阿里巴巴集团控股有限公司 项目推荐方法、装置及系统
CN111324724B (zh) * 2020-02-13 2023-04-11 腾讯科技（深圳）有限公司 数据处理方法及装置、电子设备和计算机可读存储介质
CN111522967B (zh) * 2020-04-27 2023-09-15 北京百度网讯科技有限公司 知识图谱构建方法、装置、设备以及存储介质
CN111639274B (zh) * 2020-06-01 2023-05-09 重庆锐云科技有限公司 线上商品智能排序方法、装置、计算机设备和存储介质
CN111652647B (zh) * 2020-06-03 2023-07-07 广州市丰申网络科技有限公司 一种广告自动化上新方法、系统、装置及存储介质
CN113821703B (zh) * 2020-06-18 2023-12-08 广州汽车集团股份有限公司 一种车联网用户画像生成方法及其系统
CN111768234B (zh) * 2020-06-28 2023-12-19 百度在线网络技术（北京）有限公司 为用户生成推荐文案的方法及设备、电子设备和介质
CN112084383B (zh) * 2020-09-07 2023-08-18 中国平安财产保险股份有限公司 基于知识图谱的信息推荐方法、装置、设备及存储介质
CN111931067B (zh) * 2020-09-14 2023-09-29 北京百度网讯科技有限公司 兴趣点推荐方法、装置、设备和介质
CN112241460A (zh) * 2020-10-27 2021-01-19 上海明略人工智能(集团)有限公司 辅助推荐关键词的方法、装置、电子设备及存储介质
CN112508636B (zh) * 2020-11-03 2023-01-24 上海财经大学 一种护肤品推荐方法
WO2022102975A1 (ko) * 2020-11-12 2022-05-19 삼성전자주식회사 음성 비서를 통해 광고를 제공하는 전자 장치 및 그 제어 방법
US11880866B2 (en) 2020-11-12 2024-01-23 Samsung Electronics Co., Ltd. Electronic apparatus for providing advertisement through voice assistant and control method thereof
CN112528146B (zh) * 2020-12-09 2022-05-27 北京百度网讯科技有限公司 内容资源推荐方法、装置、电子设备及存储介质
CN112380194A (zh) * 2020-12-10 2021-02-19 广东省科学院智能制造研究所 一种制造业加工设备信息图谱架构构建方法及系统
CN112685656A (zh) * 2020-12-22 2021-04-20 航天信息股份有限公司 标签推荐方法及电子设备
CN112381627B (zh) * 2021-01-14 2021-05-07 北京育学园健康管理中心有限公司 育儿知识下的商品评分处理推荐方法、装置
CN113505281B (zh) * 2021-06-15 2022-11-08 北京三快在线科技有限公司 搜索建议词生成方法、装置、电子设备及可读存储介质
JP7132448B1 (ja) 2022-01-11 2022-09-06 株式会社Zozo 提案装置、提案方法及び提案プログラム
CN114996507A (zh) * 2022-06-10 2022-09-02 北京达佳互联信息技术有限公司 视频推荐方法及装置
CN115375205B (zh) * 2022-10-25 2023-06-23 埃睿迪信息技术（北京）有限公司 一种用水用户画像的确定方法、装置及设备

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104077327A (zh) * 2013-03-29 2014-10-01 阿里巴巴集团控股有限公司 核心词重要性识别方法和设备及搜索结果排序方法和设备
CN104794232A (zh) * 2015-05-04 2015-07-22 百度在线网络技术（北京）有限公司 搜索结果的推荐方法和装置
US20160350321A1 (en) * 2012-11-28 2016-12-01 BloomReach Inc. Search with autosuggest and refinements

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP1189148A1 (en) * 2000-09-19 2002-03-20 UMA Information Technology AG Document search and analysing method and apparatus
US7565627B2 (en) 2004-09-30 2009-07-21 Microsoft Corporation Query graphs indicating related queries
US7917528B1 (en) * 2007-04-02 2011-03-29 Google Inc. Contextual display of query refinements
WO2011024585A1 (ja) 2009-08-25 2011-03-03 楽天株式会社 情報取得装置、情報取得プログラム、記録媒体、情報取得方法及び情報取得システム
US20120016642A1 (en) * 2010-07-14 2012-01-19 Yahoo! Inc. Contextual-bandit approach to personalized news article recommendation
WO2013180121A1 (ja) 2012-05-30 2013-12-05 楽天株式会社 情報処理装置、情報処理方法、情報処理プログラム及び記録媒体

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160350321A1 (en) * 2012-11-28 2016-12-01 BloomReach Inc. Search with autosuggest and refinements
CN104077327A (zh) * 2013-03-29 2014-10-01 阿里巴巴集团控股有限公司 核心词重要性识别方法和设备及搜索结果排序方法和设备
CN104794232A (zh) * 2015-05-04 2015-07-22 百度在线网络技术（北京）有限公司 搜索结果的推荐方法和装置

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP3623956A4 *

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110727859A (zh) * 2019-09-12 2020-01-24 北京十分科技有限公司 一种推荐信息推送方法及其装置
CN112288512A (zh) * 2020-10-09 2021-01-29 北京三快在线科技有限公司 信息处理方法、装置、电子设备及可读存储介质
CN112308432A (zh) * 2020-11-03 2021-02-02 重庆理工大学 一种基于类别和能力评价的众包任务推荐方法
CN113435197A (zh) * 2021-06-25 2021-09-24 平安国际智慧城市科技股份有限公司 数据推送方法、装置、推送服务器及存储介质
CN113626575A (zh) * 2021-09-01 2021-11-09 浙江力石科技股份有限公司 一种基于用户问答的智能推荐方法

Also Published As

Publication number Publication date
KR102370408B1 (ko) 2022-03-03
EP3623956A1 (en) 2020-03-18
EP3623956A4 (en) 2020-10-07
US20200117675A1 (en) 2020-04-16
JP2020523714A (ja) 2020-08-06
CN107688606A (zh) 2018-02-13
KR20200007917A (ko) 2020-01-22
CA3066941A1 (en) 2019-01-31
JP7065122B2 (ja) 2022-05-11

Similar Documents

Publication Publication Date Title
WO2019019554A1 (zh) 一种推荐信息的获取方法及装置、电子设备
US11481461B2 (en) Concept networks and systems and methods for the creation, update and use of same to select images, including the selection of images corresponding to destinations in artificial intelligence systems
US10861077B1 (en) Machine, process, and manufacture for machine learning based cross category item recommendations
US11768893B2 (en) Concept networks and systems and methods for the creation, update and use of same in artificial intelligence systems
US10198520B2 (en) Search with more like this refinements
WO2009125495A1 (ja) 広告表示方法、広告表示システム及び広告表示プログラム
CN105718184A (zh) 一种数据处理方法和装置
US20080183675A1 (en) System for updating advertisement bids
CN107229666B (zh) 一种基于推荐系统的兴趣探索方法及装置
US11138249B1 (en) Systems and methods for the creation, update and use of concept networks to select destinations in artificial intelligence systems
US10853424B1 (en) Content delivery using persona segments for multiple users
CN110532351B (zh) 推荐词展示方法、装置、设备及计算机可读存储介质
CN108897823B (zh) 基于深度学习注意力机制的个性化商品检索方法及装置
CN103136257A (zh) 信息提供方法及其装置
KR100762829B1 (ko) 소비자의 구매 취향에 따른 상품 검색 서비스 시스템 및방법
US20150160847A1 (en) System and method for searching through a graphic user interface
US9542077B2 (en) Information processing apparatus, information processing method, information processing program, and recording medium
EP4231172A1 (en) Aspect-aware autocomplete query
CN112132660B (zh) 商品推荐方法、系统、设备及存储介质
US9342834B2 (en) System and method for setting goals and modifying segment criteria counts
Shafiee et al. The effect of implementing SEO techniques and websites design methods on e-tourism development: A study of travel agencies e-tourism websites
JP6037540B1 (ja) 検索システム、検索方法およびプログラム
CN109858770A (zh) 对象质量评估方法和装置
US10296956B2 (en) Method, system, and computer-readable medium for product and vendor selection
CN110309274A (zh) 引导语推荐方法、装置及电子设备

Legal Events

Date Code Title Description
ENP Entry into the national phase

Ref document number: 3066941

Country of ref document: CA

ENP Entry into the national phase

Ref document number: 20197036939

Country of ref document: KR

Kind code of ref document: A

ENP Entry into the national phase

Ref document number: 2019569754

Country of ref document: JP

Kind code of ref document: A

ENP Entry into the national phase

Ref document number: 2017918888

Country of ref document: EP

Effective date: 20191213

NENP Non-entry into the national phase

Ref country code: DE