HK1194500A

HK1194500A - Method and device for sequencing search results

Info

Publication number: HK1194500A
Application number: HK14107659.1A
Authority: HK
Inventors: 李嘉森; 苏绥绥
Original assignee: 阿里巴巴集团控股有限公司
Filing date: 2014-07-28
Publication date: 2014-10-17

Description

Search result sorting method and device

Technical Field

The invention relates to the technical field of computers, in particular to a method and a device for sorting search results.

Background

With the continuous development of computer network technology, each e-commerce website is sequentially provided with a search engine for providing commodity search service, so that a user can conveniently inquire commodities and quickly find out commodities which the user is interested in. The commodity search is similar to the search method of a common search engine (such as hundredths, google, bin, etc.), but has the characteristics of itself. Compared with the common search, in the commodity search, in addition to considering the correlation with the query word, the commodity search also adds a plurality of dimensions such as the historical evaluation of the buyer to the commodity, the credibility, the cheating difficulty, the category correlation, the commodity price and the like of the seller who issues the commodity information, and comprehensively sorts each commodity object by combining the personal preference data of the user to obtain the search result.

Existing search results are typically presented in a paginated or waterfall stream format, with a certain number of items displayed in a page or screen, e.g., 40 items per page, and the user may choose to page back or flip through the screen. If the mode is a paging display mode, when a user needs to turn pages, the user can browse commodities on other pages by clicking the corresponding page number or sending a request to the search engine again through the backward page turning tag. If the display mode is the waterfall flow display mode, the request is sent to the search engine again when the pull-down mouse or the slider is dragged, and more webpage contents are displayed to replace the originally displayed contents.

Because the existing search results are all natural search results output at one time, new sequencing is not redone when the search results of different pages are switched. The display order of the search results has no little relation to the user's clicking or browsing behavior. For example, when a query word is input as nike during commodity search, a user clicks 10 commodities on a first page, and when a second page is viewed by page turning, a display result of the second page is irrelevant to whether the first page is clicked or not, and dynamic sequencing cannot be performed according to user behaviors.

In the existing common search engine, the first click behavior of the natural search results of the query word is used as a target webpage, and then the natural search results are adjusted according to the distance of the target webpage based on the similarity between the full amount of webpages and the webpage from small to large, so that the problems of one meaning and multiple meanings of the query word are solved, and the query intention of a user is clarified.

The distance calculation of similarity of web pages of common search is not suitable for commodity search, because information displayed on a list page of a commodity search natural result, such as title, price and picture information, is compounded by various information such as commodity description, evaluation information, shop information, deal records, sales promotion information, attribute information and the like of a target page, the information of the target page cannot represent that a user clicks the information of the natural search result, and the similarity of the results of the commodity search cannot be represented exactly by the web page similarity between the target page and an object. In addition, the dynamic sequencing of the common search mainly optimizes the natural results of the query words, detects the intention of the query words by using the initial search results, has larger deviation from the actual intention of the user, and has lower accuracy and conversion rate.

Disclosure of Invention

The invention aims to provide a method and a device for sequencing search results, which are used for identifying user intentions according to user behaviors and dynamically updating the sequencing of the search results to display the search results meeting the requirements of users, so that the search results are more accurate, the users can conveniently and quickly find required contents, and the conversion rate is high.

In order to achieve the above object, the present invention provides a method for ranking search results, the method comprising:

recording behavior information of a user on the displayed object in the search result obtained according to the query word;

when a page turning or screen turning request is received, determining the sharing degree of one or more attribute characteristics in a set of objects of which the user acts according to the behavior information of the user on the displayed objects;

selecting attribute characteristics meeting the preset requirements as reference indexes for sequencing the objects to be displayed or sequenced according to the similarity;

and adjusting the sequence of the objects to be displayed or sequenced, the attribute characteristics of which accord with the reference index.

In another aspect, the present invention further provides a device for ranking search results, where the device includes:

the recording unit is used for recording behavior information of the user on the displayed object in the search result obtained according to the query word;

the computing unit is used for determining the sharing degree of one or more attribute characteristics in the set of the objects of the behavior of the user according to the behavior information of the user on the displayed objects when receiving a page turning or screen turning request;

the selecting unit is used for selecting attribute characteristics meeting the preset requirements as reference indexes for sequencing the objects to be displayed or sequenced according to the commonness;

and the adjusting unit is used for adjusting the sequence of the objects to be displayed or sequenced, the attribute characteristics of which accord with the reference indexes selected by the selecting unit.

According to the search result ordering method and device provided by the invention, the intention of the user is identified according to the latest browsing or clicking behavior of the user, the ordering of the search results is dynamically updated to display the search results meeting the requirements of the user, so that the search results are more accurate, the user can quickly find the required content, and the conversion rate is high.

Drawings

Fig. 1 is a flowchart of a method for ranking search results according to an embodiment of the present invention;

fig. 2 is a flowchart of a method for determining the sharing degree of the attribute features according to the behavior information of the user on the displayed object according to an embodiment of the present invention;

FIG. 3 is a flowchart of a method for calculating a ranking score of an object to be displayed or ranked according to an embodiment of the present invention;

fig. 4 is a schematic diagram of a search result sorting apparatus according to a second embodiment of the present invention.

Detailed Description

The technical solution of the present invention is further described in detail by the accompanying drawings and embodiments.

The method and the device for sorting the search results can be used in various search engines, particularly under the condition that the attribute information of the obtained search results has more types, such as scenes of commodity search of an e-commerce website and the like.

Example one

Fig. 1 is a flowchart of a method for sorting search results provided in this embodiment, and as shown in fig. 1, the method for sorting search results of the present invention includes:

and step S101, recording behavior information of the displayed object in the search result obtained according to the query word by the user.

After the user inputs the query words through the browser and confirms the query words, the user initiates a search request to a search engine. The search engine receives a search request of a user, performs processing operations such as word segmentation on the query word, retrieves to obtain a corresponding search result, and displays the search result in a paging or screen segmentation mode according to the degree of correlation. In this step, the search result of the query term may be obtained by using an existing search engine and displayed by using a default ranking method.

The object may be a commodity or commodity information. In the commodity search, the query term is used for searching out the commodities related to the query term from the database, and the commodities are arranged and displayed according to the degree of the relevance, and the arrangement mode of the commodities can also adopt other modes.

User behavior on objects in search results includes browsing and clicking. And recording the behavior information of each object in the search result according to behaviors of the user such as browsing or clicking. Behavioral information may include, but is not limited to: the user behavior object, the position information of the user behavior object in the search result page and/or the sequence of browsing or clicking the behavior object by the user. In this embodiment, a product search is taken as an example for explanation, and one object in the search result is specifically a product. And the objects which are not shown in the search result are the objects to be reordered and shown.

According to the browsing behavior of the user, recording the information of the page browsed by the user, and recording the page as long as the page is browsed by the user no matter whether the page generates click information or not. The recorded information, that is, the user behavior information further includes session information of the page, a query term corresponding to the page, a page number or a screen of the page in the search result, a commodity identification code (id) corresponding to the page, and a behavior sequence of an object of the current behavior of the user in each object of the behavior of the user.

Session refers to a process of communication between an end user and an interactive system, and generally refers to a process from user registration to system logout.

Specifically, a session refers to a period of time from when a user enters a website to when the browser is closed when the user browses the website, and all operations of the user using the browser in the period of time are operations in the same session.

In the same session, a user may input one or more query terms to perform search query, in this embodiment, search results under the same query term or a combination of query terms in the same session are dynamically ranked, and search results on the (n + 1) th page are ranked according to browsing or click feedback of the first n pages. Of course, under the actual use requirement, the search results of the same query term of different sessions in a period of time may also be dynamically ranked, for example, for the same user or the same IP address.

Similarly, according to the clicking behavior of the user, the commodities clicked by the user are recorded, and the order of the clicked commodities, namely the behavior order of the user on the object, is recorded together. For an object clicked by a user, the recorded behavior information includes session information of a page corresponding to the object, a query word corresponding to the object, a page number or a screen of a page where the object is located in a search result, a commodity identification code (id) of the object, and a click sequence of the object in each object of the user behavior. The format of the record may be (session, query word, page number, commodity id string, behavior sequence), which represents the information that the commodity id in a certain page number is clicked under a certain query word of a session. Statistics shows that the last click of the user in the query search is the one closest to the user intention, but not the first click. Generally, the later in the click sequence, the more valuable it is to the user's query intent.

For the storage format, clicked items may be written into the item id string in sequence, compressed with a fixed symbol, and a data structure of a "stack" may be used, but is not limited.

For example, if "nike" is searched, 4 items are clicked in sequence on the first page, and the number of items on a general page is 40, the result is recorded as follows:

(20120324081，nike，1，auction_1：auction_3：auction_5：auction_15)

the first field represents a session, such as 20120324081 is the id of the session, the second field represents the query term, and the third field represents the page. action _1, action _3, action _5, and action _15 represent different article ids, respectively. According to the record of the above example, it can be seen that, in the search result displayed on the first page, the item whose item id is animation _15 is clicked fourth.

And S102, when a page turning or screen turning request is received, determining the sharing degree of one or more attribute characteristics in the set of the objects of the behavior of the user according to the behavior information of the user on the displayed objects.

When a user initiates a page turning or screen turning request aiming at a search result list of the same query word in the same session, the search engine acquires the behavior information of the user on the displayed object under the search result corresponding to the query word used by the current session, and determines the characteristic information of the user behavior object according to the behavior information of the user. As shown in fig. 2, the method for determining the degree of commonality of one or more preset attribute features in a set of objects in which a user has acted according to behavior information of the user on the displayed objects specifically includes the following sub-steps S102_1 to S102_ 4.

And step S102_1, acquiring the attribute characteristics of the displayed object.

And sequentially analyzing the commodity id corresponding to the user behavior object according to the related sequence in the history selection information, and acquiring various attribute characteristics corresponding to the commodity id. The attribute characteristics comprise: the information of the commodity comprises one or more of a title of commodity information, a price of the commodity, a picture or picture address of the commodity, a recent transaction number, a freight rate, a region where the commodity is located, a name of a seller and a custom label (for example, a service label provided by a commodity publisher, such as a false claim three, which is described in detail, a commodity or a seller label with no reason for returning goods for 7 days, lightning delivery, a detail drawing, payment for goods arrival, consumer security and the like). . Since the attribute features of the commodities can be usually displayed on a list page of the search results, the behavior tendency of the buyer to the search results can be intuitively influenced, and therefore the intention of the user can be identified by using the commodity attribute features. The attribute features comprise attributes and attribute values or attribute value intervals of the commodities on the attributes.

And S102_2, classifying the objects clicked by the user into the selected set, and classifying the objects which are not clicked in the displayed objects into the unselected set.

The user click comprises the operation behavior of clicking to enter a detail page of a certain object by selecting the object in a search result page. The user click may also be an operation behavior included in the search result page in which the user selects an object to expand the detailed description information.

When a user selects a certain list page of search results, the objects presented on the list page are typically objects that the user can retrieve or browse. After the user selects to obtain the list page, the object displayed on the list page is regarded as the object browsed by the user.

All viewed items (i.e., displayed objects) are divided into two sets: the method comprises the steps of selecting a set and an unselected set, wherein the commodities in the selected set are browsed and clicked commodities, and the commodities in the unselected set are browsed and clicked commodities.

It should be noted that the order of step S102_1 and step S102_2 may be exchanged.

And step S102_3, calculating the sharing degree of each attribute feature in the selected set according to the attribute features of the objects in the selected set.

And S102_4, calculating the sharing degree of the attribute characteristics in the unselected set according to the attribute characteristics of the objects in the unselected set.

The commonality of a certain attribute feature in the selected set or the unselected set is specifically: in the selected set or the unselected set, a ratio of the number of the objects having the same or similar attribute characteristics to the total number of the objects in the selected set or the unselected set, that is, a ratio of the number of the objects having the same or similar attribute values on the attribute corresponding to the attribute characteristics to the total number of the objects in the selected set or the unselected set.

The case where the attribute values of the attributes are similar includes: the attribute values of the objects on the attribute are in the same preset interval.

Since the picture is one of the main factors affecting the click, but the pictures of each commodity hardly have the degree of commonality, the attribute features are classified into picture attribute features and non-picture attribute features. The attribute value of the picture attribute may be represented using a feature value of a picture of the commodity.

For the calculation of the commonality of the non-picture attribute features, the attribute values of the non-picture attribute features of the product are digitized or group-discretized. For example, it can use statistical method to group the attributes of price, number of deals, credit, etc. according to a certain rule, for example, the attribute of price can be divided into three attribute value intervals of (0,50], (50, 100) and (100 + 150), etc., and the attribute value of each commodity on the price attribute is divided into corresponding intervals.

Continuing to refer to fig. 1, step S103, selecting the attribute features meeting the predetermined requirements as reference indexes for ordering the objects to be displayed or sorted according to the commonality.

The method for selecting the attribute characteristics meeting the predetermined requirements as the reference index for sorting the objects to be displayed or sorted (i.e. objects not displayed) may include one or more of the following ways:

and sequencing all the attribute characteristics in sequence from large to small according to the commonalities, and selecting a preset number of attribute characteristics sequenced in the front as reference indexes.

Or, the attribute features with the commonness larger than the set threshold are used as reference indexes.

Or, calculating the difference of the degree of commonality of each attribute feature in the selected set and the unselected set, and using the corresponding attribute feature as a reference index when the difference of the degree of commonality is greater than a set threshold value.

Or, the difference of the degrees of commonality of each attribute feature in the selected set and the unselected set is calculated first, the attribute features are sorted from large to small according to the difference of the degrees of commonality, and a predetermined number of the previously sorted attribute features are selected as reference indexes.

If one attribute feature has higher degree of commonality in the selected set, namely the attribute feature in the commodity set clicked by the user has higher degree of commonality, the attribute feature represents a reference factor for selecting the commodity by the user when the attribute feature is likely to be selected; meanwhile, if the lower the commonality of the attribute feature in the unselected set is, the larger the difference between the commonalities of the attribute feature in the selected set and the unselected set is, the higher the possibility that the attribute feature affects the selected commodity of the user is, the attribute feature can be used as a reference index for dynamic sorting of commodities in the non-displayed page. It will be appreciated that the reference indicators may be used to distinguish user intent.

The dynamic sorting of this embodiment may comprehensively consider the difference between the degrees of commonality of the attribute features in the selected set and the unselected set, and select the attribute with high degree of commonality in the selected set and without the attribute feature or with low degree of commonality in the unselected set as the reference index.

It should be noted that step S102_2 is not an essential step, and when the number of clicks made by the user is small, for example, only 1-2 items are clicked, the unselected set may not be considered. Accordingly, step S102_1 may sequentially acquire each attribute of the clicked object and the attribute value thereof according to the click order of the object only.

And step S104, calculating respective sorting scores of the objects to be displayed or sorted according to the reference indexes.

And determining the reference index according to the commonalities. For example, the set of reference indicators determined according to the degree of commonality of the attribute features in the selected set is denoted as (a 1, a2, A3). The set of reference indicators determined by the degree of commonality of the respective attribute features in the unselected set is denoted (B1, B2, B3).

For the objects to be sorted, sorting score adjustment is carried out by using the reference indexes, the objects meeting the reference indexes A1, A2 and/or A3 are subjected to bonus, and the objects meeting the reference indexes B1, B2 and/or B3 are subjected to down-grading; that is, objects that are not shown or to be sorted and that meet the reference index determined from the selected set are given a positive ranking score or are given a positive ranking score, and objects that are not shown or to be sorted and that meet the reference index determined from the selected set are given a negative ranking score or are given a negative ranking score.

Calculating the respective sorting scores of the objects to be displayed or sorted can simultaneously consider the sorting score of each object obtained according to the natural sorting rule, namely, according to the reference index, adding or subtracting the sorting score obtained according to the natural sorting rule. The natural ranking rule may be a ranking rule adopted when a home page or home screen of the search result list is displayed.

It can be understood that, in order to ensure that the objects meeting the reference index determined from the selected set can be preferentially displayed, when the respective ranking scores of the objects to be displayed or ranked are calculated, the ranking score of the objects not displayed or ranked meeting the reference index determined from the selected set may also be a score obtained by adding points on the basis of the maximum ranking score of the objects not displayed or ranked according to the natural ranking rule.

In addition, considering the influence of the selection sequence of the user on the displayed objects on the user' S intention, step S104 includes steps S104_1-S104_4 in the specific embodiment of the present application, as shown in fig. 3. The order of selection by the user on the presented objects is the order in which the user clicks on the objects in the presented search result list.

And step S104_1, endowing a first weight to the object clicked by the user in the displayed objects according to the click sequence.

Since the value of the click sequence is higher and more suitable for the query intention of the user, the clicked item is given a different first weight according to the click sequence, for example, the rank of the sequence may be directly used as the first weight, and if the item is clicked first, the first weight of the item is V₁And the first weight of the second clicked commodity is V₂By analogy, the first weight of the nth clicked commodity is V_n. According to the order of the clicked commodities, the first weight of each clicked commodity is increased in sequence, namely V₁<V₂…<V_n。

Step S104_2, calculating the weight of each reference index according to the weight of the object meeting the reference index in the objects selected by the user.

The number of objects selected by the user (i.e., clicked objects) is n, which is a natural number. The weight of the influence of the objects selected by the user on the query intention of the user is V in sequence₁、V₂、…V_n. For a reference index, the sum of the weights of the selected objects matching the reference index is calculated and denoted V_m. Of selected objects that are to meet the referenceThe ratio of the sum of the weights to the sum of the weights of the objects selected by the user is taken as the weight of the reference index, i.e., the weight of the influence of the reference index on the user query intention, and is denoted as Q. I.e., Q = V_m/(V₁+V₂+…V_n)。

For example, the user sequentially clicks 1 st, 3 th, 5 th and 15 th commodities with respect to the first page search result with the query term nike, the records are (20120324081, nike, 1 st, operation _ 1: operation _ 3: operation _ 5: operation _15), and the first weights corresponding to the 1 st, 3 th, 5 th and 15 th commodities are 1, 2, 3 and 4, respectively. The reference index determined according to the user behavior includes the attribute of the label related to the product and the corresponding attribute value "false claim three". The 1 st, 5 th and 15 th commodities in the search result list have the attribute of a commodity-related label, and the commodity-related label has the attribute value of 'false one claim three', namely, the reference index is met. The weight of the reference index is (1 +3+ 4)/(1 +2+3+ 4) = 0.8.

For the calculation of the picture characteristics, the pictures of the commodities can be converted into the characteristic values of the pictures under the offline condition, the pictures are represented in a numerical form, and the closer the numerical values are, the more similar the pictures are.

And step S104_3, calculating the comprehensive score of the influence degree of the reference index which is accorded with the object to be displayed or ordered on the ordering of the object.

The object to be sorted may conform to a plurality of reference indexes, and index scores may be respectively preconfigured for each reference index. The index of the reference index determined from the selected set is classified as a positive value, such as a score of "1"; the index of the reference index determined from the unselected set is classified as a negative value, such as a score of "-1". And the comprehensive score of the influence degree of the reference indexes accorded by the objects to be sorted on the sorting is the sum of the products of the indexes of each reference index and the weight.

The calculation formula of the comprehensive score of the influence degree of the reference index which the object to be sorted accords with on the object to be sorted is as follows:

f（auction_id）=T₁*Q₁+T₂*Q₂…+T_n*Q_n

wherein f (animation _ id) represents the comprehensive score of the influence degree of the reference index which the object to be sorted accords with on the sorting of the object to be sorted. T is_nIs the index score of the nth reference index, Q_nIs the weight of the nth reference index. For the reference index determined according to the selected set, the index is divided into fixed positive values, and can also be different preset values respectively. For the reference index determined according to the unselected set, the index may be a fixed negative value, or may be different preset negative values.

And for each click picture, calculating the similarity between the picture of the commodity to be sorted and the click picture according to the characteristic value of the picture, setting a picture similarity threshold, judging whether the calculated similarity is greater than a preset picture similarity threshold, and selecting the first N3 pictures, wherein N3 is any positive integer. For each page displaying 40 commodities, N3 may also be set to 40, and the commodities with the pictures of the commodities to be sorted and the similarity of the clicked pictures ranked in the top 40 are selected and stored as (animation _ id, auc _ list). Wherein auc _ list is a list of items, arranged in order of magnitude of similarity, the first item appearing in auc _ list is most similar to the clicked item in picture feature.

And if the clicking behavior of the user comprises a plurality of clicked commodities, determining a commodity list of commodities to be sorted for each clicked commodity. And if one commodity to be sorted exists in a certain commodity list, weighting the similarity of the commodity to be sorted and the clicked commodity into f (animation _ id). If not, no weighting is applied.

And step S104_4, adjusting the sorting score of the object to be displayed or sorted according to the comprehensive score of the influence degree of the reference index which is met by the object to be displayed or sorted on the object sorting.

When f (animation _ id) is a positive value, the sorting score of the object to be displayed or sorted by the reference index is adjusted to be a forward adjustment, namely, the object sorting is promoted.

When f (animation _ id) is a negative value, the adjustment of the sorting score of the object to be displayed or sorted by the reference index is negative adjustment, namely the sorting of the object is reduced.

And S105, reordering the objects to be displayed or ordered according to the adjusted ordering scores of the objects to be displayed or ordered.

If f (animation _ id) <0, indicating that the commodities are to be sorted in a descending order, adding f (animation _ id) to the original sorting score of the commodities to obtain an adjusted sorting score; if f (animation _ id) >0, the commodities are represented to be sorted, and f (animation _ id) is added on the basis of the original sorting score of the commodities to obtain the adjusted sorting score.

Therefore, the commodities to be sorted can be dynamically adjusted according to the behavior characteristics of the user, and the search results of the (n + 1) th page are sorted according to the click feedback of the first n pages.

Optionally, when the ranking score is adjusted, browsing behavior or click feedback behavior of the user in a period of time may also be added, such as which commodities are browsed, which shops, which categories, and which commodities are collected, and common attribute features are extracted based on statistics of historical preference to dynamically influence the ranking score.

The above is a detailed description of the method for sorting search results provided by the present invention, and the following is a detailed description of the apparatus for sorting search results provided by the present invention.

Example two

Fig. 4 is a schematic diagram of a sorting apparatus for search results provided in this embodiment, and as shown in fig. 4, the sorting apparatus of the present invention includes: the device comprises a recording unit 10, a calculating unit 20, a selecting unit 30, an adjusting unit 40, a classifying unit 50 and a sorting unit 60.

The recording unit 10 is used for recording behavior information of a user on a displayed object in a search result obtained according to the query term.

After the user inputs the query words through the browser and confirms the query words, the user initiates a search request to a search engine. The search engine receives a search request of a user, performs processing operations such as word segmentation on the query word, retrieves to obtain a corresponding search result, and displays the search result in a paging or screen segmentation mode according to the degree of correlation. The search results of the query terms can be obtained by utilizing the existing search engine and displayed by adopting a default sorting method.

The object may be a commodity or commodity information. In the commodity search, the retrieval unit 10 retrieves the commodities related to the query term from the database by using the query term, and arranges and displays the commodities according to the degree of the relevance, the arrangement mode of the commodities may be, and the specific display form is not limited in the present invention.

User behavior on objects in search results includes browsing and clicking. The recording unit 10 records behavior information of each object in the search result according to behaviors of a user such as browsing or clicking, and when receiving a page turning or screen turning request of the same query word, triggers the calculating unit 20. Behavioral information may include, but is not limited to: the user behavior object, the position information of the user behavior object in the search result page and/or the sequence of browsing or clicking the behavior object by the user. In this embodiment, a product search is taken as an example for explanation, and one object in the search result is specifically a product. And the objects which are not shown in the search result are the objects to be reordered and shown.

The recording unit 10 records information of a page browsed by a user according to browsing behavior of the user, and records the page as long as the page is browsed by the user regardless of whether click information is generated on the page or not. The recorded information, that is, the user behavior information further includes session information of the page, a query term corresponding to the page, a page number or a screen of the page in the search result, a commodity identification code (id) corresponding to the page, and a behavior sequence of an object of the current behavior of the user in each object of the behavior of the user.

Similarly, the recording unit 10 records the commodities clicked by the user according to the clicking behavior of the user, and also records the order in which the commodities are clicked, that is, the behavior order of the user for the object. For an object clicked by a user, the recorded behavior information includes session information of a page corresponding to the object, a query word corresponding to the object, a page number or a screen of a page where the object is located in a search result, a commodity identification code (id) of the object, and a click sequence of the object in each object of the user behavior. The format of the record may be (session, query word, page number, commodity id string, behavior sequence), which represents the information that the commodity id in a certain page number is clicked under a certain query word of a session. Statistics shows that the last click of the user in the query search is the one closest to the user intention, but not the first click. Generally, the later in the click sequence, the more valuable it is to the user's query intent.

For the storage format, the recording unit 10 may write the clicked goods in the goods id string in sequence, compress with a fixed symbol, and may use, but is not limited to, a data structure of "stack".

(20120324081，nike，1，auction_1：auction_3：auction_5：auction_15)

When the recording unit 10 receives a page turning or screen turning request initiated by the user for the same query word in the same session, the computing unit 20 is triggered.

The computing unit 20 is configured to determine, according to the behavior information of the user on the displayed object, a degree of commonality of the one or more attribute features in the set of objects where the user has acted.

When triggered by the recording unit 10, the calculating unit 20 obtains behavior information of the user on the displayed object under the search result corresponding to the query word used by the current session, and determines feature information of the user behavior according to the behavior information of the user.

The classifying unit 50 is connected to the calculating unit 20, and is configured to classify the objects clicked by the user into the selected set, and classify the objects not clicked in the displayed objects into the unselected set.

Of course, the classifying unit 50 may be used only for classifying the objects clicked on in the search result into the selected set, or may be used only for classifying the objects that are not clicked on and are shown in the search result into the unselected set.

The calculation unit 20 includes an acquisition subunit 201, a first calculation subunit 202, and a second calculation subunit 203.

The acquiring subunit 201 is configured to acquire attribute features of the exposed object.

The first calculating subunit 202 is configured to calculate a degree of commonality of each attribute feature in the selected set according to the recorded behavior information of the user on the displayed object.

The second calculating subunit 203 is configured to calculate, according to the attribute features of the objects in the unselected set, a degree of commonality of the attribute features in the unselected set.

The selecting unit 30 is configured to select, according to the commonality, an attribute feature meeting a predetermined requirement as a reference index for ordering the objects to be displayed or ordered.

The method for selecting the attribute characteristics meeting the predetermined requirement by the selecting unit 30 as the reference index for sorting the objects to be displayed or sorted (i.e. objects not displayed) may include one or more of the following ways:

It should be noted that the calculating unit 20 may not be provided with the classifying unit 50, and when the number of clicks of the user is small, for example, only 1-2 items are clicked, the unselected set may not be considered. Accordingly, the first calculating subunit 202 is adopted to perform calculation, and the respective degrees of commonality of the attribute features are calculated according to the attribute features or the corresponding attribute feature values of all the selected objects. The selecting unit 30 takes the attribute feature whose commonality satisfies the requirement as the key feature.

The sorting unit 40 is configured to calculate respective sorting scores of the objects to be displayed or sorted according to the reference index.

The selection unit 30 determines the reference index according to the degree of commonality. For example, the set of reference indicators determined according to the degree of commonality of the attribute features in the selected set is denoted as (a 1, a2, A3). The set of reference indicators determined by the degree of commonality of the respective attribute features in the unselected set is denoted (B1, B2, B3).

The sorting unit 40 performs sorting score adjustment on the objects to be sorted by using the reference indexes, performs bonus score on the objects meeting the reference indexes A1, A2 and/or A3, and performs reduction score on the objects meeting the reference indexes B1, B2 and/or B3; that is, objects that are not shown or to be sorted and that meet the reference index determined from the selected set are given a positive ranking score or are given a positive ranking score, and objects that are not shown or to be sorted and that meet the reference index determined from the selected set are given a negative ranking score or are given a negative ranking score.

The sorting unit 40 may calculate the respective sorting scores of the objects to be displayed or sorted, and may simultaneously consider the sorting score of each object obtained according to the natural sorting rule, that is, the sorting score obtained according to the natural sorting rule is added or subtracted according to the reference index. The natural ranking rule may be a ranking rule adopted when a home page or home screen of the search result list is displayed.

It can be understood that, in order to ensure that the objects meeting the reference index determined from the selected set can be preferentially displayed, when the sorting unit 40 calculates the respective sorting scores of the objects to be displayed or sorted, the sorting score of the object not displayed or sorted meeting the reference index determined from the selected set may also be a score obtained by adding scores based on the maximum sorting score of the objects not displayed or sorted according to the natural sorting rule.

In addition, in consideration of the influence of the selection sequence of the user on the displayed objects on the user's intention, the sorting unit 40 in the embodiment of the present application further includes: an assignment subunit 401, a third calculation subunit 402, a fourth calculation subunit 403 and a fifth calculation subunit 404. The order of selection by the user on the presented objects is the order in which the user clicks on the objects in the presented search result list.

The assignment subunit 401 is configured to assign a first weight to an object clicked by the user in the displayed objects according to the click sequence.

The third calculating subunit 402 is configured to calculate a weight of each reference indicator according to the first weight of the object meeting the reference indicator in the objects selected by the user.

The number of objects selected by the user (i.e., clicked objects) is n, which is a natural number. The weight of the influence of the objects selected by the user on the query intention of the user is V in sequence₁、V₂、…V_n. For a reference index, the sum of the weights of the selected objects matching the reference index is calculated and denoted V_m. The ratio of the sum of the weights of the selected objects that meet the reference index to the sum of the weights of the objects selected by the user is taken as the weight of the reference index, i.e., the weight of the influence of the reference index on the query intention of the user, and is denoted as Q. I.e., Q = V_m/(V₁+V₂+…V_n)。

For example, the user sequentially clicks 1 st, 3 th, 5 th and 15 th commodities with respect to the first page search result with the query term nike, the records are (20120324081, nike, 1 st, operation _ 1: operation _ 3: operation _ 5: operation _15), and the first weights corresponding to the 1 st, 3 th, 5 th and 15 th commodities are 1, 2, 3 and 4, respectively. The third calculating subunit 402 determines a reference index according to the user behavior, which includes the attribute of the label associated with the product and the corresponding attribute value "false claim three". The 1 st, 5 th and 15 th commodities in the search result list have the attribute of a commodity-related label, and the commodity-related label has the attribute value of 'false one claim three', namely, the reference index is met. The weight of the reference index is (1 +3+ 4)/(1 +2+3+ 4) = 0.8.

For the calculation of the picture features, the third calculation subunit 402 selects to convert the picture of the commodity into the feature value of the picture under the offline condition, and represents the picture in a numerical form, wherein the closer the numerical value is, the more similar the picture is.

The fourth calculating subunit 403 is configured to calculate, according to the weight of the reference indicator to which the attribute feature of the object to be displayed or sorted conforms, a comprehensive score of the degree of influence of the reference indicator on the object sorting.

The fourth calculating subunit 403 adopts the following calculation formula of the comprehensive score of the influence degree of the reference index that the object to be sorted conforms to on the object to be sorted:

f（auction_id）=T₁*Q₁+T₂*Q₂…+T_n*Q_n

For each clicked picture, the fourth calculating subunit 403 calculates the similarity between the picture of the commodity to be sorted and the clicked picture according to the feature value of the picture, sets a picture similarity threshold, determines whether the calculated similarity is greater than a preset picture similarity threshold, and selects the first N3 pictures, where N3 is any positive integer. For each page displaying 40 commodities, N3 may also be set to 40, and the commodities with the pictures of the commodities to be sorted and the similarity of the clicked pictures ranked in the top 40 are selected and stored as (animation _ id, auc _ list). Wherein auc _ list is a list of items, arranged in order of magnitude of similarity, the first item appearing in auc _ list is most similar to the clicked item in picture feature.

If the user's click behavior includes a plurality of clicked items, the fourth calculating subunit 403 determines a item list of items to be sorted for each clicked item. And if one commodity to be sorted exists in a certain commodity list, weighting the similarity of the commodity to be sorted and the clicked commodity into f (animation _ id). If not, no weighting is applied.

The fifth calculating subunit 404 is configured to adjust the ranking score of the object according to the composite score.

The fifth calculating subunit 404 adjusts the ranking score of the object to be displayed or to be ranked according to the comprehensive score of the degree of influence of the reference index to which the object to be displayed or to be ranked conforms to the ranking of the object.

The sorting unit 60 is configured to reorder the objects to be displayed or to be sorted according to the adjusted sorting scores of the objects to be displayed or to be sorted.

The sorting method and the sorting device provided by the invention can dynamically update the search sorting of the subsequent pages according to the click feedback behavior in front of the comprehensive sorting so as to optimize the sorting of the search results of the query words, meet the requirements of users, and enable the users to quickly find the required content, thereby improving the conversion rate from browsing to bargaining of the users.

Those of skill would further appreciate that the various illustrative components and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or combinations of both, and that the various illustrative components and steps have been described above generally in terms of their functionality in order to clearly illustrate this interchangeability of hardware and software. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the implementation. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.

The steps of a method or algorithm described in connection with the embodiments disclosed herein may be embodied in hardware, a software module executed by a processor, or a combination of the two. A software module may reside in Random Access Memory (RAM), memory, Read Only Memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

The above-mentioned embodiments are intended to illustrate the objects, technical solutions and advantages of the present invention in further detail, and it should be understood that the above-mentioned embodiments are merely exemplary embodiments of the present invention, and are not intended to limit the scope of the present invention, and any modifications, equivalent substitutions, improvements and the like made within the spirit and principle of the present invention should be included in the scope of the present invention.

Claims

1. A method for ranking search results, the method comprising:

2. The method of ranking search results according to claim 1, wherein the behavior information includes: the user behavior object, the position information of the user behavior object in the search result page and/or the sequence of browsing or clicking the behavior object by the user.

3. The method of ranking search results according to claim 1, wherein the object is a commodity or commodity information.

4. The method for sorting search results according to claim 1, wherein the method for determining the commonality of one or more attribute features in the set of objects in which the user acts according to the behavior information of the user on the displayed objects comprises:

acquiring the attribute characteristics of the displayed object;

classifying the objects clicked by the user into a selected set, and classifying the objects which are not clicked in the displayed objects into an unselected set;

calculating the degree of commonality of each attribute feature in the selected set according to the recorded behavior information of the user on the displayed object; and

and calculating the sharing degree of the attribute features in the unselected set according to the attribute features of the objects in the unselected set.

5. The method of claim 4, wherein the degree of commonality of attribute features in the selected or unselected set is: a ratio of the number of objects in the selected or unselected set having the same or similar attribute characteristic to the total number of objects in the selected or unselected set.

6. The method according to claim 4, wherein the selecting, according to the commonality, an attribute feature meeting a predetermined requirement as a reference index for ranking the objects to be displayed or ranked specifically comprises:

and sequencing all the attribute features in sequence from large to small according to the degree of commonality, and selecting a preset number of the attribute features sequenced at the front as a reference index or selecting the attribute features with the degree of commonality larger than a set threshold value as the reference index.

7. The method according to claim 4, wherein the selecting, according to the commonality, an attribute feature meeting a predetermined requirement as a reference index for ranking the objects to be displayed or ranked specifically comprises:

calculating the difference of the degrees of similarity of the various attribute features in the selected set and the unselected set, sorting the various attribute features from large to small according to the difference of the degrees of similarity, and selecting a preset number of previously sorted attribute features as reference indexes or selecting the corresponding attribute features as the reference indexes when the difference of the degrees of similarity is larger than a set threshold value.

8. The method of claim 1, wherein the object of the user behavior is clicked in the search result.

9. The method of ranking search results according to claim 8, further comprising:

classifying the clicked object in the search result into the selected set;

the method for adjusting the sequence of the objects to be displayed or sequenced, the attribute characteristics of which accord with the reference index, comprises the following steps:

calculating the degree of commonality of each attribute feature in the selected set according to the behavior information of the user on the objects in the selected set;

selecting the attribute characteristics with the similarity greater than a preset threshold value as reference indexes;

and improving the sequence of the objects to be displayed or sequenced, the attribute characteristics of which accord with the reference index.

10. The method of ranking search results according to claim 8, further comprising:

classifying the clicked object in the search result into the selected set;

sorting all attribute features in sequence from large to small according to the commonalities, and selecting a preset number of previously sorted attribute features as reference indexes;

11. The method of ranking search results according to claim 1, further comprising:

classifying objects which are shown in the search results and are not clicked into an unselected set;

calculating the sharing degree of each attribute feature in the unselected set according to the behavior information of the user on the objects in the unselected set;

and descending the sequence of the objects to be displayed or sequenced, the attribute characteristics of which accord with the reference index.

12. The method of ranking search results according to claim 1, further comprising:

13. The method of ranking search results according to claim 1, further comprising:

calculating respective sorting scores of the objects to be displayed or sorted according to the reference indexes;

the step of adjusting the sequence of the objects to be displayed or sequenced, the attribute characteristics of which conform to the reference index, is specifically as follows: reordering the objects to be displayed or ordered according to the ordering score;

calculating respective sorting scores of the objects to be displayed or sorted according to the reference indexes, specifically comprising:

according to the click sequence, giving a first weight to an object clicked by a user in the displayed object;

calculating the weight of each reference index according to the first weight of the object meeting the reference index in the objects selected by the user;

calculating the comprehensive score of the influence degree of the reference index on the object sorting according to the weight of the reference index which is accorded with the attribute characteristics of the object to be displayed or sorted; and

and adjusting the sorting score of the object according to the comprehensive score.

14. An apparatus for ranking search results, the apparatus comprising:

15. The apparatus for ranking search results according to claim 14, wherein the behavior information includes: the user behavior object, the position information of the user behavior object in the search result page and/or the sequence of browsing or clicking the behavior object by the user.

16. The apparatus for sorting search results according to claim 14, wherein the object is a commodity or commodity information.

17. The apparatus for ranking search results according to claim 14, wherein said apparatus further comprises:

the classification unit is used for classifying the objects clicked by the user into the selected set and classifying the objects which are not clicked in the displayed objects into the unselected set;

the calculation unit specifically includes:

the acquisition subunit is used for acquiring the attribute characteristics of the displayed object;

the first calculating subunit is used for calculating the sharing degree of each attribute feature in the selected set according to the recorded behavior information of the user on the displayed object; and

and the second calculating subunit is used for calculating the sharing degree of the attribute features in the unselected set according to the attribute features of the objects in the unselected set.

18. The apparatus of claim 17, wherein the degree of commonality of the attribute features in the selected or unselected sets is: a ratio of the number of objects in the selected or unselected set having the same or similar attribute characteristic to the total number of objects in the selected or unselected set.

19. The apparatus for ranking search results according to claim 17, wherein the selecting unit is specifically configured to:

20. The apparatus for ranking search results according to claim 17, wherein the selecting unit is specifically configured to:

21. The apparatus for sorting search results according to claim 14, wherein the object of the user behavior is clicked in the search result.

22. The apparatus for ranking search results according to claim 21, wherein said apparatus further comprises:

the classification unit is used for classifying the clicked objects in the search results into a selected set;

the computing unit is specifically used for computing the sharing degree of each attribute feature in the selected set according to the behavior information of the user on the objects in the selected set;

the selecting unit is specifically configured to select the attribute feature with the commonality larger than a preset threshold as a reference index;

the adjusting unit is specifically configured to promote the ranking of the objects to be displayed or ranked, of which the attribute characteristics conform to the reference index.

23. The apparatus for ranking search results according to claim 21, wherein said apparatus further comprises:

the selecting unit is specifically used for sequentially sorting all attribute features from large to small according to the commonness, and selecting a preset number of previously sorted attribute features as reference indexes;

24. The apparatus for ranking search results according to claim 14, wherein said apparatus further comprises:

the classification unit is used for classifying the objects which are shown in the search results and are not clicked into an unselected set;

the computing unit is specifically used for computing the sharing degree of each attribute feature in the unselected set according to the behavior information of the user on the objects in the unselected set;

the adjusting unit is specifically configured to sort the objects to be displayed or sorted whose attribute characteristics meet the reference index in a descending order.

25. The apparatus for ranking search results according to claim 14, wherein said apparatus further comprises:

26. The apparatus for ranking search results according to claim 14, wherein said apparatus further comprises:

the sorting unit is used for calculating respective sorting scores of the objects to be displayed or sorted according to the reference indexes selected by the selecting unit;

the adjusting unit reorders the objects to be displayed or to be ordered according to the ordering scores calculated by the ordering unit;

the sorting unit specifically includes:

the assignment subunit is used for assigning a first weight to the object clicked by the user in the displayed objects according to the click sequence;

the third calculation subunit is used for calculating the weight of each reference index according to the first weight of the object which meets the reference index in the objects selected by the user;

the fourth calculating subunit is used for calculating the comprehensive score of the influence degree of the reference index on the object sorting according to the weight of the reference index which is accorded with the attribute characteristics of the object to be displayed or sorted; and

and the fifth calculating subunit is used for adjusting the sorting score of the object according to the comprehensive score.