CN102810117A - A method and device for providing search results - Google Patents

Abstract

The invention aims to provide a method and equipment for supplying a search result. Computer equipment acquires initial search results corresponding to a query sequence input by a user; a first sequencing model is used for screening out preferable search results from the initial search results; a second sequencing model is used for screening out the optimal search results from the preferable search results; and the optimal search results are supplied to the user. Compared with the prior art, the method and the equipment have the advantages that the search results are screened according to different levels through a plurality of sequencing models, so that the optimal search results are supplied to the user; and therefore, the precision and the efficiency of the search results are considered, and the search efficiency and the search effect are optimized. Moreover, the sequencing models are determined by a machine learning method; an upper layer model is generated by a submodel through the machine learning; and therefore, the arrangement of the sequencing models is optimized; and the instantaneity, the understandability and the controllability of the sequencing models of the systems are guaranteed.

Description

A method and device for providing search results

technical field

The present invention relates to the field of computers, in particular to a technology for providing search results.

Background technique

At present, most of the search results are provided in a one-time sorting method, that is, according to the user's query request, through the query of the background database, using the preset sorting model, the search result corresponding to the user's query request is provided to the user. There are certain problems in this method, that is, it is not easy to achieve the optimization of efficiency and effect in one sorting. From the perspective of efficiency priority, query methods with low precision can be used, but the high accuracy of query results cannot be guaranteed; from the perspective of effect priority, query methods with high precision can be used, but query efficiency cannot be guaranteed at the same time.

Contents of the invention

The object of the present invention is to provide a method and device for providing search results.

According to one aspect of the present invention, there is provided a computer-implemented method for providing search results, the method comprising the steps of:

a obtaining initial search results corresponding to the query sequence entered by the user;

b using the first ranking model to filter out preferred search results from the initial search results;

c using the second sorting model to filter out the optimal search result from the preferred search results;

d providing said optimal search result to said user.

According to another aspect of the present invention, there is also provided a result providing device for providing search results, the device comprising:

A result obtaining device, configured to obtain an initial search result corresponding to the query sequence input by the user;

A first screening device, configured to use a first sorting model to filter out preferred search results from the initial search results;

The second screening means is used to use the second sorting model to screen out the optimal search result from the preferred search results;

The result providing means is used for providing the optimal search result to the user.

According to still another aspect of the present invention, a search engine is also provided, including the above-mentioned result providing device for providing search results.

According to still another aspect of the present invention, a search engine plug-in is also provided, including the above-mentioned result providing device for providing search results.

According to still another aspect of the present invention, a browser is also provided, including the above-mentioned result providing device for providing search results.

According to still another aspect of the present invention, a browser plug-in is also provided, including the above-mentioned result providing device for providing search results.

Compared with the prior art, the present invention classifies and screens the search results through multiple sorting models, realizes providing the optimal search results to the user, thereby taking into account the accuracy and efficiency of the search results, and achieving the optimization of search efficiency and effect . Furthermore, the machine learning method is used to determine the ranking model, and the sub-model is used to generate the upper model through machine learning, thereby optimizing the setting of the ranking model and ensuring the real-time performance, comprehensibility and controllability of the system ranking model. In addition, most of the current feature selection for ranking models is to put all the underlying features from various angles together. It is not conducive to the reuse of previous rule system accumulation; in view of this, the present invention also uses different features for different sorting models, while ensuring the optimization of search efficiency and effect, optimizes the feature selection of the sorting model, further improves the search efficiency and Effect optimization.

Description of drawings

Other characteristics, objects and advantages of the present invention will become more apparent by reading the detailed description of non-limiting embodiments made with reference to the following drawings:

FIG. 1 shows a schematic diagram of a result providing device for providing search results according to one aspect of the present invention;

Fig. 2 shows a schematic diagram of a result providing device for providing search results according to a preferred embodiment of the present invention;

FIG. 3 shows a flowchart of a method for providing search results implemented by a result providing device according to another aspect of the present invention;

Fig. 4 shows a flowchart of a method for providing search results implemented by a result providing device according to a preferred embodiment of the present invention.

The same or similar reference numerals in the drawings represent the same or similar components.

Detailed ways

The present invention will be described in further detail below in conjunction with the accompanying drawings.

Fig. 1 shows a schematic diagram of a result providing device for providing search results according to one aspect of the present invention; wherein, the result providing device includes a result obtaining device 11, a first screening device 12, a second screening device 13, a result providing device 14. The result obtaining means 11 obtains the initial search result corresponding to the query sequence input by the user; the first screening means 12 uses the first sorting model to filter out the preferred search results in the initial search results; the second filtering means 13 uses the second A ranking model, selecting the optimal search result from the preferred search results; the result providing means 14 provides the optimal search result to the user. Wherein, the result providing device can not only work independently, but can also be integrated with network equipment, user equipment, or a device formed by integrating network equipment and user equipment through a network. Wherein, the network device includes but is not limited to a computer, a network host, a single network server, a plurality of network server sets or a cloud formed by multiple servers; here, the cloud consists of a large number of computers or networks based on cloud computing (Cloud Computing) Server configuration, among them, cloud computing is a kind of distributed computing, a virtual supercomputer composed of a group of loosely coupled computer sets. The user equipment includes but is not limited to any electronic product that can interact with the user through a keyboard, remote control, touch pad, or voice-activated device, such as a computer, smart phone, PDA, game console, or IPTV. The network includes, but is not limited to, the Internet, a wide area network, a metropolitan area network, a local area network, a VPN network, a wireless ad hoc network (AdHoc network) and the like. Those skilled in the art should understand that other result-providing devices are also applicable to the present invention, and should also be included in the protection scope of the present invention, and are incorporated herein by reference.

Wherein, the result obtaining means 11 obtains the initial search result corresponding to the query sequence input by the user. Specifically, the result acquisition means 11, for example, uses page technology, such as JSP, ASP, PHP and other page technologies, or by calling the application program interface (API) provided by the user equipment or other equipment that can provide the query sequence, or http , https and other agreed communication methods, interact with users, obtain the query sequence input by the user, and obtain The initial search results corresponding to the query sequence input by the user, wherein the user can interact with the result acquisition device 11 through a keyboard, touch screen, or voice input device, and input the query sequence he wants to query, thereby initiating a search; or, the result acquisition The device 11 is based on various communication protocols (Communications Protocol), where "communication protocol" refers to the transmission protocol of computer communication, such as: TCP/IP, UDP, FTP, ICMP, NetBEUI, etc., and also includes other communication protocols existing in the computer. form of communication, such as: communication between objects in object-oriented programming; message transfer protocols between different programs in an operating system or between different modules of a computer, interacting with other devices that can provide said initial search results, such as search engines, to Get the initial search results corresponding to the query sequence entered by the user. Preferably, the result obtaining means 11 may also intercept a certain number of search results from the obtained search results corresponding to the query sequence input by the user, as the initial search results. For example, the user proposes a query sequence of "the best Sichuan cuisine" to the result acquisition device 11 through page technology, and the result acquisition device 11 performs word segmentation on "the best Sichuan cuisine", and separates "the best Sichuan cuisine" and "Sichuan cuisine" in the database. "Retrieve and obtain 1000 initial search results. Those skilled in the art should understand that the above-mentioned methods of obtaining initial search results and several communication transmission protocols are only examples, and other existing or future methods of obtaining initial search results or communication transmission protocols that are applicable to the present invention are also applicable. It should be included within the protection scope of the present invention, and is included here by reference.

The first screening means 12 uses the first sorting model to filter out preferred search results from the initial search results. Specifically, the first screening means 12 uses the first ranking model to calculate the priority or ranking information of each initial search result for the initial search results provided by the result obtaining device 11; information, and screen the initial search results to obtain the preferred search results, for example, select the initial search results whose priority or ranking information meets a certain threshold as the preferred search results, or select these initial search results according to their priority or The sorting information is arranged in descending order, and the top N initial search results are used as the preferred search results. Wherein, the first sorting model includes but is not limited to a sorting algorithm, a sorting feature vector, and the like.

Here, the method of using the first ranking model to calculate the priority or ranking information of the initial search results includes but not limited to: using the first ranking model, for example, a feature vector including one or more feature components and their weights, to determine the initial search result The assignment of the corresponding feature components, so as to obtain the feature vector corresponding to the initial search result, that is, the feature vector includes the assignment and weight of the feature components, as the priority or sorting information of the initial search result; preferably Alternatively, the assignment of the feature vector may also be weighted and determined according to the assignment and weight of each feature component included in the feature vector, so as to serve as the priority or ranking information of the initial search result.

Here, the ways of arranging the initial search results according to their priority or ranking information include but not limited to: without generality, it may be assumed that the priority or ranking information of the initial search results includes feature vectors corresponding to the initial search results, The eigenvector includes the assignment of one or more eigencomponents and their weights, and the ordering of the corresponding search results can be determined according to the size of the assignment of each eigenvector (as determined by the weight of the assignment of its eigencomponents); or, according to The weight of each feature component of each feature vector and its assignment size (such as dictionary sorting) to determine the ordering of the corresponding search results, for example, first sort by the assignment of the feature component with the highest weight, and then for the feature with the highest weight The initial search results with the same component assignment can be sorted according to the assignment of the feature component with the second highest weight until the sorting of all initial search results is completed. For example, the user proposes a query sequence of "the best Sichuan cuisine" to the result acquisition device 11 through page technology, and the result acquisition device 11 provides 1000 initial search results. Sorting is performed based on two components with a weight of 50% each, authoritative analysis and semantic analysis, and the first screening device 12 obtains the 1000 initial search results provided by the result obtaining device 11, and utilizes the authoritative and semantic ranking The method screens the 1000 items, and obtains the top 100 results sorted by semantics and authority as the preferred search results.

The second screening means 13 uses the second sorting model to screen the optimal search results from the preferred search results. Specifically, the second screening device 13 uses the second sorting model to further screen the user's query sequence in the preferred search results provided by the first screening device 12 to obtain the optimal search results. result. Those skilled in the art should be able to understand that, except for the difference between the second sorting model and the first sorting model, the implementation of the second screening device 13 is the same as or basically similar to that of the first screening device 12. incorporated herein by reference. For example, the user proposes a query sequence of "the best Sichuan cuisine" to the result acquisition device 11 through page technology, the result acquisition device 11 provides 1000 initial search results, the first screening device 12 provides 100 preferred search results, and the second The sorting model is defined as performing sorting on the preferred results of the user query sequence based on four components with a weight of 25% each: user demand analysis, user behavior statistical result analysis, authoritative analysis, and semantic analysis. The second screening device 13 obtains the first A screening device 12 provides 100 preferred search results, and utilizes the second sorting model to carry out comprehensive sorting of 4 components to the 100 preferred search results to obtain further sorting results, and use the top 10 results as The best search results.

The result providing means 14 provides the optimal search result to the user. Specifically, the result providing means 14 acquires the optimal search results screened by the second screening means 13, and utilizes the user interaction, or according to the application program interface (API) provided by the user equipment or http, https, etc. The format of the means of communication requires that the optimal search results be provided to the user. For example, the 10 optimal search results obtained by the second screening device 13 are presented to the user as the first page of the user's search results, or the 10 optimal search results obtained by the second screening device 13 are used as the user's homepage. The first page of the search results is presented to the user, and the remaining preferred search results are presented to the user sequentially after the first page as required. Here, those skilled in the art should be able to understand that the display of the initial search result, the preferred search result or the optimal search result may be performed sequentially according to the optimal search result, the preferred search result, and the initial search result; Choose one of the optimal search results, preferred search results, and initial search results to display; or combine two or three of the optimal search results, preferred search results, and initial search results to display according to user needs of.

Here, those skilled in the art should understand that the result providing device may also include a third screening device or even more screening devices, so as to perform multi-level sorting on the initial search results, for example, adopt a multi-level sorting model to sort the initial search results step by step , to obtain the optimal search results to be provided to users. Preferably, the present invention can also determine the level of the required sorting model according to different application requirements, and according to the multi-level sorting model of the corresponding level, such as a secondary sorting model, a three-level or more sorting model, the initial search results Sort step by step to obtain the optimal search result to be provided to the user.

Preferably, the first screening device 12 can also use the first ranking model to determine the priority of the initial search results; according to a predetermined first quantity threshold, based on the priority of the initial search results, from the initial The preferred search results are screened out from the search results, wherein the number of the preferred search results satisfies the first number threshold. Specifically, the first screening device 12 uses the first sorting model to determine the priority of the initial search result, such as the feature vector corresponding to the initial search result or its assignment, and then according to the predetermined first quantity threshold, based on The priority of these initial search results, screening out the preferred search results from the initial search results, wherein the number of the preferred search results meets the first number threshold, for example, directly screening out the initial search results A certain number of initial search results with higher priorities are used as the preferred search results, or these initial search results are first arranged in descending order according to their priorities, and then a certain number of initial search results ranked in the top are used as the preferred search results result. For example, assuming that the first quantity threshold is 100, the user puts forward the query sequence of "the best Sichuan cuisine" to the result acquisition device 11 through page technology, and the result acquisition device 11 provides 1000 initial search results, and the first ranking model is defined as The initial search results of the user query sequence are sorted based on two components with a weight of 50%, authoritative analysis and semantic analysis, and the first screening device 12 obtains the 1000 initial search results provided by the result acquisition device 11, And use the authoritative and semantic sorting method to determine the priority of the 1000 initial search results, and according to the first quantity threshold of 100, filter the 1000 initial search results in descending order of their priority to obtain the semantic and The top 100 results sorted authoritatively are used as the preferred search results.

More preferably, the result providing device further includes a first threshold determining means (not shown), wherein the first threshold determining means determines the first quantity threshold according to a predetermined first quantity determining rule; wherein the first The number determination rule includes at least any one of the following: based on the number of the initial search results, determining the first number threshold; based on a predetermined determination rule for determining the number threshold of the optimal search result, determining the first A quantitative threshold. Specifically, the first threshold determining means dynamically determines the first number threshold according to the number of the initial search results or a predetermined determination rule for determining the number threshold of the optimal search results. For example, if the first quantity threshold is set in the first quantity determination rule to be proportional to the quantity of the initial search results, the greater the quantity of the initial search results, the greater the first quantity threshold; or, the first quantity The first number threshold set in the determination rule is positively related to the determination rule of the number threshold of the optimal search result. If the limit of the number threshold of the optimal search result is fixed according to the limit displayed on the interface, then the first number threshold It is directly proportional to the number threshold of the fixed optimal search result, for example, according to the number of impressions per page, the first number threshold is determined to be an integer multiple of the number of impressions per page; or, the determination of the first number threshold is consistent with the initial search The number of results forms a positive feedback relationship with a predetermined determination rule for determining the number threshold of the optimal search result, and finally reaches a balance, thereby determining the first number threshold.

Preferably, the second screening means 13 can also use the second ranking model to determine the priority of the preferred search results; according to a predetermined second quantity threshold, based on the priority of the preferred search results, from the preferred The optimal search result is selected from the search results, wherein the number of the optimal search results satisfies the second number threshold. Specifically, the second screening device 13 uses the second sorting model to determine the priority of the preferred search result, such as the feature vector corresponding to the preferred search result or its assignment, and then according to the predetermined second quantity threshold, based on The priority of these preferred search results, the optimal search results are selected from the preferred search results, wherein the number of the optimal search results meets the second number threshold, for example, directly from these preferred search results Select a certain number of preferred search results with higher priority as the optimal search results, or first arrange these preferred search results in descending order according to their priorities, and then use a certain number of preferred search results at the top as the optimal search results. the best search results.

More preferably, the result providing device further includes a second threshold determining means (not shown), wherein the second threshold determining means determines the second quantity threshold according to a predetermined second quantity determining rule; wherein the second The quantity determination rule includes at least any one of the following: determining the second quantity threshold based on the terminal attribute of the user equipment of the user; determining the second quantity threshold based on the type information of the query sequence; determining the second quantity threshold based on the preferred The number of search results determines the second number threshold. Specifically, the second threshold determining means is based on the predetermined second quantity determining rule, based on the terminal attribute of the user equipment of the user, or based on the type information of the query sequence, or based on the number of the preferred search results, to determine the second quantity threshold. For example, according to the different terminal attributes of the user equipment, the second number threshold is also different accordingly. For example, the display screen of the PC terminal is relatively large, and each page can present 10 results, then the second number threshold is 10, and the display screen of the mobile device is relatively large. Small, then the second quantity threshold is 6; or according to the type information of the query sequence, the second quantity threshold is also correspondingly different, if the type of the query sequence is train number information, then the most accurate small amount of results can be presented, the first The second quantity threshold is relatively small. If the type of the query sequence is catering information, relatively more results may be presented to meet the user's needs; or it is directly proportional to the quantity of the preferred search results to determine the second quantity threshold.

Figure 2 shows a schematic diagram of a result providing device for providing search results according to a preferred embodiment of the present invention; wherein, the result providing device includes a result obtaining device 11', a first screening device 12', and a second screening device 13 ', result providing means 14', model determining means 15'. Specifically, the model determination device 15' determines the ranking model by means of machine learning according to the labeled and sorted first search result training data, wherein the ranking model includes at least any one of the following: the first ranking model, The second sorting model; the result obtaining means 11' obtains initial search results corresponding to the query sequence input by the user; the first screening means 12' utilizes the first sorting model to filter out preferred search results from the initial search results The second screening means 13' utilizes the second sorting model to filter out the optimal search result from the preferred search results; the result providing means 14' provides the optimal search result to the user. Among them, the result obtaining device 11', the first screening device 12', the second screening device 13', and the result providing device 14' in the result providing device are respectively the same or basically the same as the corresponding devices shown in Figure 1, so no further details are given here. , and is incorporated herein by reference.

The above-mentioned devices are continuously working. Here, those skilled in the art should understand that "continuous" means that the above-mentioned devices respectively perform the determination of the sorting model and the initial search results according to the set or real-time adjusted working mode requirements. Obtaining, screening of preferred search results, screening of optimal search results, provision of optimal search results, etc., until the result providing device stops obtaining the initial search results corresponding to the query sequence input by the user.

The model determining device 15' determines a ranking model by means of machine learning according to the labeled and sorted first search result training data, wherein the ranking model includes at least any one of the following: the first ranking model, the first Two sorting models. Specifically, the model determination device 15' uses the initial ranking model or a random feature component as the initial ranking model according to the first search result training data that has been marked and sorted, and continuously uses a linear model, a nonlinear model, or a combination thereof as required. Adjust the parameters in the ranking model, and determine the ranking model by means of machine learning, such as the first ranking model or the second ranking model. Wherein, the first search result training data includes but is not limited to query strings (query) and corresponding search results (url), each query string is marked with a correlation level number with the search result, or has a correlation The multiple search results distinguished by high and low indicate the high and low relationships among the multiple search results. Wherein, if the ranking model is a linear model, the ranking model includes but not limited to each feature component and the weight corresponding to the feature component; if the ranking model is a nonlinear model, the ranking The model can include, for example, a decision threshold corresponding to a certain threshold point of a feature component. The entire ranking model is composed of several decision thresholds, for example, a decision tree is formed by decision thresholds of multiple feature components, and then multiple decision trees are used to form a sorting Model for comprehensive scoring of search results. For example, the first search result training data is continuously substituted into the current ranking model, such as the initial ranking model or the learned intermediate ranking model, and the ranking information of the training data is calculated, such as the weighted sum of multiple feature components with weight information , or the score obtained by scoring one or more decision trees composed of multiple feature components with decision thresholds, and adjust the current ranking model according to the difference between the ranking information and the marked ranking information, such as adding Subtracting feature components of the current ranking model or adjusting weight information or decision thresholds of feature components, for example, adjusting weight information or decision thresholds of multiple feature components sequentially or simultaneously. Those skilled in the art should be able to understand that, here, determining the ranking model through machine learning not only makes the error of the ranking model on the training data of the first search result as small as possible, but also has a certain generalization ability.

Preferably, the first search result training data includes but is not limited to query sequences, search results, and the mapping relationship between query sequences and search results, such as the priority, ranking or score of the search results under the corresponding query sequence etc.; preferably, the mapping relationship between the query sequence and the search result also includes the weight or decision threshold of the feature component of the search result under the corresponding query sequence. For example, for a given model prototype, such as a feature vector containing multiple feature components, but the parameters of each feature component have not been calibrated, such as the weight or decision threshold of the feature component, use the query sequence and search The training set of the mapping relationship between the results, through genetic algorithm, neural network, decision tree, support vector machine and other machine learning algorithms, determine the model parameters including the parameters corresponding to each feature component, that is, obtain the ranking model, as described The first ranking model or the second ranking model.

Preferably, the result providing device further includes sub-model determining means 16', wherein the sub-model determining means 16' determines one or more of An ordering submodel of the feature components in the ordering model. Those skilled in the art should be able to understand that, except for the difference between the ranking sub-model and the ranking model, the implementation of the sub-model determining device 16' is the same as or basically similar to that of the model determining device 15', so for the sake of brevity, no further details are given, and only reference method is included here. Wherein, the second search result training data includes but not limited to query sequence, search result, and the mapping relationship between query sequence and search result, such as the priority, ranking or score of the search result under the corresponding query sequence, etc. ; Preferably, the mapping relationship between the query sequence and the search result also includes the weight or decision threshold of the feature component of the search result under the corresponding query sequence. For example, for a given sub-model prototype, such as a feature vector containing multiple feature components, but the parameters of each feature component have not been calibrated, such as the weight or decision threshold of the feature component, use the query sequence and The training set of the mapping relationship between search results, through machine learning algorithms such as genetic algorithm, neural network, decision tree, support vector machine, etc., determine the model parameters including the parameters corresponding to each feature component, that is, obtain the sub-ranking model.

Preferably, the model determining means 15' can also determine the second ranking model by means of machine learning according to the labeled and sorted training data of the first search results, wherein the second ranking model includes user behavior feature components. Specifically, the user behavior characteristic information includes but not limited to the time information of the user's search, the address information confirmed according to the user's IP address, the operation information about the search results generated by the user through clicking, touching, swiping the screen, and the time spent on the page, etc. The model determination device 15' can use the first search result training data marked and sorted, and use machine learning algorithms such as genetic algorithms and neural networks to perform machine learning on the user behavior characteristic information, and determine the user behavior characteristics under different user behavior characteristic information. The weights or decision thresholds of the feature components, i.e. to obtain the second ranking model. Here, the user behavior feature information may be included in the first search result training data, or may be stored in a third-party device such as a search engine or a search log database connected to the result-providing device via a network, and through such third-party The application program interface (API) provided by the device obtains the user behavior feature information from the third-party devices.

In another preferred embodiment, the above-mentioned result providing device for providing search results can be combined with an existing search engine to form a new search engine, and the existing search engine can be, for example, Google's Google Search engine, baidu search engine of baidu company, etc.

In another preferred embodiment, the above-mentioned result providing device for providing search results can be combined with an existing search engine plug-in to form a new search engine plug-in, and the existing search engine plug-in can be, for example, Google The company's Google ToolBar, Baidu's Baidu Sobar, Microsoft's MSN ToolBar, etc.

In another preferred embodiment, the above-mentioned result providing device for providing search results can be combined with an existing browser to form a new browser, and the existing browser can be, for example, the IE browser of Microsoft Corporation , Netscape browser of Netscape Company, Firefox browser of Mozilla Company, Chrome browser of Google Company, Maxthon browser of Aoyou Company, opera browser of Opera Company, 360 browser of 360 Company, Sogou browser of Sohu Company, Tencent's Tencent TT browser, etc.

In another preferred embodiment, the above-mentioned result providing device for providing search results can be combined with an existing browser plug-in to form a new browser plug-in, and the existing browser plug-in can be, for example, Flash Plug-ins, RealPlayer plug-ins, MMS plug-ins, MIDI staff plug-ins, ActiveX plug-ins, etc.

Fig. 3 shows a flow chart of a method for providing search results implemented by a result providing device according to another aspect of the present invention; specifically, in step s1, the result providing device obtains the search results corresponding to the query sequence input by the user Initial search results; in step s2, use the first sorting model to filter out the preferred search results from the initial search results; in step s3, use the second sorting model to filter out the best search results in the preferred search results Search results; in step s4, the optimal search results are provided to the user. Wherein, the result providing device can not only work independently, but can also be integrated with network equipment, user equipment, or a device formed by integrating network equipment and user equipment through a network. Wherein, the network device includes but is not limited to a computer, a network host, a single network server, a plurality of network server sets or a cloud formed by multiple servers; here, the cloud consists of a large number of computers or networks based on cloud computing (Cloud Computing) Server configuration, among them, cloud computing is a kind of distributed computing, a virtual supercomputer composed of a group of loosely coupled computer sets. The user equipment includes but is not limited to any electronic product that can interact with the user through a keyboard, remote control, touch pad, or voice-activated device, such as a computer, smart phone, PDA, game console, or IPTV. The network includes, but is not limited to, the Internet, a wide area network, a metropolitan area network, a local area network, a VPN network, a wireless ad hoc network (AdHoc network) and the like. Those skilled in the art should understand that other result-providing devices are also applicable to the present invention, and should also be included in the protection scope of the present invention, and are incorporated herein by reference.

Wherein, in step s1, the result providing device obtains the initial search result corresponding to the query sequence input by the user. Specifically, in step s1, the result providing device, for example, uses page technology, such as JSP, ASP, PHP and other page technologies, or by calling the application programming interface (API) provided by the user equipment or other equipment that can provide the query sequence. ), or other agreed communication methods such as http, https, etc. to interact with the user, obtain the query sequence input by the user, perform word segmentation on the query sequence input by the user, and search for the query sequence after word segmentation in the query database Obtain the initial search results corresponding to the query sequence input by the user in a manner, wherein the user can interact with the result providing device through a keyboard, touch screen, or voice input device, and input the query sequence he wants to query, thereby initiating a search; or , the results provided by the device are based on various communication protocols (Communications Protocol), where "communication protocol" refers to the transmission protocol of computer communication, such as: TCP/IP, UDP, FTP, ICMP, NetBEUI, etc., and also includes the information existing in the computer Other forms of communication, such as: communication between objects in object-oriented programming; message transfer protocol between different programs in the operating system or different modules of the computer, and other devices that can provide the initial search results, such as search engines, conduct Interact to obtain initial search results corresponding to the query sequence entered by the user. Preferably, in step s1, the result providing device may also intercept a certain number of search results from the acquired search results corresponding to the query sequence input by the user, as the initial search results. For example, the user puts forward a query sequence of "the best Sichuan food" to the result providing device through page technology, and the result providing device performs word segmentation for "the best Sichuan food", and separates "the best Sichuan food" and "Sichuan food" in the database. Retrieval, obtained 1000 initial search results. Those skilled in the art should understand that the above-mentioned methods of obtaining initial search results and several communication transmission protocols are only examples, and other existing or future methods of obtaining initial search results or communication transmission protocols that are applicable to the present invention are also applicable. It should be included within the protection scope of the present invention, and is included here by reference.

In step s2, the result providing device uses the first ranking model to filter out preferred search results from the initial search results. Specifically, in step s2, the result providing device uses the first ranking model to calculate the priority or ranking information of each initial search result among the initial search results provided in step s1; and then according to The priority or ranking information screens the initial search results to obtain the preferred search results, for example, the initial search results whose priority or ranking information meets a certain threshold are used as the preferred search results, or these initial search results Arrange in descending order according to their priority or sorting information, and use the top N initial search results as the preferred search results. Wherein, the first sorting model includes but is not limited to a sorting algorithm, a sorting feature vector, and the like.

Here, the method of using the first ranking model to calculate the priority or ranking information of the initial search results includes but not limited to: using the first ranking model, for example, a feature vector including one or more feature components and their weights, to determine the initial search result The assignment of the corresponding feature components, so as to obtain the feature vector corresponding to the initial search result, that is, the feature vector includes the assignment and weight of the feature components, as the priority or sorting information of the initial search result; preferably Alternatively, the assignment of the feature vector may also be weighted and determined according to the assignment and weight of each feature component included in the feature vector, so as to serve as the priority or ranking information of the initial search result.

Here, the ways of arranging the initial search results according to their priority or ranking information include but not limited to: without generality, it may be assumed that the priority or ranking information of the initial search results includes feature vectors corresponding to the initial search results, The eigenvector includes the assignment of one or more eigencomponents and their weights, and the ordering of the corresponding search results can be determined according to the size of the assignment of each eigenvector (as determined by the weight of the assignment of its eigencomponents); or, according to The weight of each feature component of each feature vector and its assignment size (such as dictionary sorting) to determine the ordering of the corresponding search results, for example, first sort by the assignment of the feature component with the highest weight, and then for the feature with the highest weight The initial search results with the same component assignment can be sorted according to the assignment of the feature component with the second highest weight until the sorting of all initial search results is completed.

For example, the user proposes a query sequence of "the best Sichuan cuisine" to the result providing device through page technology, and the result providing device obtains 1000 initial search results in step s1, and the first sorting model is defined as the initial search for the user query sequence The results are sorted based on two components with a weight of 50% each, authoritative analysis and semantic analysis, and the result providing device uses the method of authoritative and semantic sorting based on the 1000 initial search results obtained in step s1 The 1000 items are screened to obtain the top 100 results sorted by semantics and authority as the preferred search results.

In step s3, the result providing device selects the optimal search result from the preferred search results by using the second ranking model. Specifically, in step s3, the result providing device uses the second ranking model to further filter the query sequence of the user from the preferred search results provided in step s2, so as to obtain the optimal search result result. Those skilled in the art should be able to understand that, except for the difference between the second sorting model and the first sorting model, the implementation of step s3 is the same as or basically similar to step s2, so for the sake of brevity, it will not be repeated, and it is only included in the reference. this. For example, the user proposes a query sequence of "the best Sichuan cuisine" to the result providing device through page technology, the result providing device provides 1000 initial search results in step s1, and the result providing device provides 100 preferred search results in step s2 As a result, the second sorting model is defined as sorting the preferred results of the user query sequence based on four components with a weight of 25% each: user demand analysis, user behavior statistical result analysis, authoritative analysis and semantic analysis, in step s3 Among them, the result providing device obtains the 100 preferred search results provided by the result providing device in step s2, and uses the second sorting model to perform comprehensive sorting of the 100 preferred search results with 4 components to obtain further sorting results, And the top 10 results are used as the optimal search results.

In step s4, the result providing device provides the optimal search result to the user. Specifically, in step s4, the result providing device obtains the optimal search result screened by the result providing device in step s3, and interacts with the user, or according to the application program interface (API) or http , https and other agreed upon format requirements for communication, and provide the optimal search results to the user. For example, the 10 optimal search results obtained by the result providing device in step s3 are presented to the user as the first page of the user's search results, or the 10 optimal search results obtained in step s3 are used as the user's The first page of the search results is presented to the user, and the remaining preferred search results are presented to the user sequentially after the first page as required. Here, those skilled in the art should be able to understand that the display of the initial search result, the preferred search result or the optimal search result may be performed sequentially according to the optimal search result, the preferred search result, and the initial search result; Choose one of the optimal search results, preferred search results, and initial search results to display; or combine two or three of the optimal search results, preferred search results, and initial search results to display according to user needs of.

Here, those skilled in the art should understand that this exemplary method may also include other steps for screening the intermediate search results or even more screening steps, so as to perform multi-level ranking on the initial search results, for example, using a multi-level ranking model to The initial search results are sorted step by step to obtain the optimal search results to be provided to the user. Preferably, the present invention can also determine the level of the required sorting model according to different application requirements, and according to the multi-level sorting model of the corresponding level, such as a secondary sorting model, a three-level or more sorting model, the initial search results Sort step by step to obtain the optimal search result to be provided to the user.

Preferably, in step s2, the result providing device can also use the first ranking model to determine the priority of the initial search results; according to a predetermined first quantity threshold, based on the priority of the initial search results, from The preferred search results are screened out from the initial search results, wherein the number of the preferred search results satisfies the first number threshold. Specifically, in step s2, the result providing device uses the first ranking model to determine the priority of the initial search result, such as the feature vector corresponding to the initial search result or its assignment, and then according to the predetermined first number Threshold, based on the priority of these initial search results, filter out the preferred search results from the initial search results, wherein the number of the preferred search results satisfies the first number threshold, for example, from the initial search results Directly filter out a certain number of initial search results with higher priority as the preferred search results, or first arrange these initial search results in descending order according to their priorities, and then use a certain number of initial search results at the top as the preferred search results. the preferred search results described above. For example, assuming that the first quantity threshold is 100, the user proposes a query sequence of "the best Sichuan cuisine" to the result providing device through page technology, and the result providing device provides 1000 initial search results in step s1, the first sorting model defines In order to sort the initial search results of the user query sequence based on two components with a weight of 50% each, authoritative analysis and semantic analysis, in step s2, the result providing device obtains the 1000 items it provided in step s1 The initial search results, and use authoritative and semantic ranking methods to determine the priority of these 1000 initial search results, and according to the first quantity threshold of 100, in the sequence of these 1000 initial search results in descending order of their priority Screening is performed to obtain the top 100 results sorted by semantics and authority as the preferred search results.

More preferably, this embodiment further includes step s7 (not shown), wherein, in step s7, the result providing device determines the first quantity threshold according to a predetermined first quantity determination rule; wherein, the first The number determination rule includes at least any one of the following: based on the number of the initial search results, determining the first number threshold; based on a predetermined determination rule for determining the number threshold of the optimal search result, determining the first A quantitative threshold. Specifically, in step s7, the result providing device dynamically determines the first number threshold according to the number of the initial search results or a predetermined determination rule for determining the number threshold of the optimal search result. For example, if the first quantity threshold is set in the first quantity determination rule to be proportional to the quantity of the initial search results, the greater the quantity of the initial search results, the greater the first quantity threshold; or, the first quantity The first number threshold set in the determination rule is positively related to the determination rule of the number threshold of the optimal search result. If the limit of the number threshold of the optimal search result is fixed according to the limit displayed on the interface, then the first number threshold It is directly proportional to the number threshold of the fixed optimal search result, for example, according to the number of impressions per page, the first number threshold is determined to be an integer multiple of the number of impressions per page; or, the determination of the first number threshold is consistent with the initial search The number of results forms a positive feedback relationship with a predetermined determination rule for determining the number threshold of the optimal search result, and finally reaches a balance, thereby determining the first number threshold.

Preferably, in step s3, the result providing device can also use the second ranking model to determine the priority of the preferred search result; according to a predetermined second quantity threshold, based on the priority of the preferred search result, from The optimal search result is selected from the preferred search results, wherein the number of the optimal search results satisfies the second number threshold. Specifically, in step s3, the result providing device uses the second ranking model to determine the priority of the preferred search result, such as the feature vector corresponding to the preferred search result or its assignment, and then according to the predetermined second number threshold, and based on the priority of these preferred search results, the optimal search results are selected from the preferred search results, wherein the number of the optimal search results satisfies the second number threshold, for example, from these preferred search results A certain number of preferred search results with higher priority are directly selected from the results as the optimal search results, or these preferred search results are first sorted in descending order according to their priorities, and then a certain number of preferred search results ranked in the front The result is used as the optimal search result.

More preferably, this embodiment further includes step s8 (not shown), wherein, in step s8, the result providing device determines the second quantity threshold according to a predetermined second quantity determination rule; wherein, the second The quantity determination rule includes at least any one of the following: determining the second quantity threshold based on the terminal attribute of the user equipment of the user; determining the second quantity threshold based on the type information of the query sequence; determining the second quantity threshold based on the preferred The number of search results determines the second number threshold. Specifically, in step s8, the result providing device determines the rule according to the predetermined second quantity, based on the terminal attribute of the user equipment of the user, or based on the type information of the query sequence, or based on the preferred search result quantity to determine the second quantity threshold. For example, according to the different terminal attributes of the user equipment, the second number threshold is also different accordingly. For example, the display screen of the PC terminal is relatively large, and each page can present 10 results, then the second number threshold is 10, and the display screen of the mobile device is relatively large. Small, then the second quantity threshold is 6; or according to the type information of the query sequence, the second quantity threshold is also correspondingly different, if the type of the query sequence is train number information, then the most accurate small amount of results can be presented, the first The second quantity threshold is relatively small. If the type of the query sequence is catering information, relatively more results may be presented to meet the user's needs; or it is directly proportional to the quantity of the preferred search results to determine the second quantity threshold.

Fig. 4 shows a flow chart of a method for providing search results implemented by a result providing device according to a preferred embodiment of the present invention; specifically, in step s5', the result providing device ranks according to the marked first search As a result of the training data, a ranking model is determined by means of machine learning, wherein the ranking model includes at least any one of the following: the first ranking model, the second ranking model; in step s1', the result providing device Obtain an initial search result corresponding to the query sequence input by the user; in step s2', the result providing device uses the first sorting model to filter out preferred search results from the initial search result; in step s3', the result provides The device uses the second ranking model to filter out the optimal search result from the preferred search results; in step s4', the result providing device provides the optimal search result to the user. Wherein, step s1', step s2', step s3' and step s4' in the result providing device are respectively the same or basically the same as the corresponding steps shown in Figure 3, so they will not be repeated here, and are included here by reference.

The above-mentioned steps are continuously working. Here, those skilled in the art should understand that "continuous" means that the above-mentioned steps respectively carry out the determination of the sorting model and the initial search results according to the requirements of the set or real-time adjusted working mode. Obtaining, screening of preferred search results, screening of optimal search results, provision of optimal search results, etc., until the result providing device stops obtaining the initial search results corresponding to the query sequence input by the user.

In step s5', the result providing device determines a ranking model through machine learning according to the labeled and sorted first search result training data, wherein the ranking model includes at least any one of the following: the first ranking model , the second ranking model. Specifically, step s5' uses the initial ranking model or any feature component as the initial ranking model based on the first search result training data that has been marked and sorted, and continuously adjusts the ranking by using a linear model, a nonlinear model, or a combination thereof as required The parameters in the model determine the ranking model, such as the first ranking model or the second ranking model, through machine learning. Wherein, the first search result training data includes but is not limited to query strings (query) and corresponding search results (url), each query string is marked with a correlation level number with the search result, or has a correlation The multiple search results distinguished by high and low indicate the high and low relationships among the multiple search results. Wherein, if the ranking model is a linear model, the ranking model includes but not limited to each feature component and the weight corresponding to the feature component; if the ranking model is a nonlinear model, the ranking The model can include, for example, a decision threshold corresponding to a certain threshold point of a feature component. The entire ranking model is composed of several decision thresholds, for example, a decision tree is formed by decision thresholds of multiple feature components, and then multiple decision trees are used to form a sorting Model for comprehensive scoring of search results. For example, the first search result training data is continuously substituted into the current ranking model, such as the initial ranking model or the learned intermediate ranking model, and the ranking information of the training data is calculated, such as the weighted sum of multiple feature components with weight information , or the score obtained by scoring one or more decision trees composed of multiple feature components with decision thresholds, and adjust the current ranking model according to the difference between the ranking information and the marked ranking information, such as adding Subtracting feature components of the current ranking model or adjusting weight information or decision thresholds of feature components, for example, adjusting weight information or decision thresholds of multiple feature components sequentially or simultaneously. Those skilled in the art should be able to understand that, here, determining the ranking model through machine learning not only makes the error of the ranking model on the training data of the first search result as small as possible, but also has a certain generalization ability.

Preferably, the first search result training data includes but is not limited to query sequences, search results, and the mapping relationship between query sequences and search results, such as the priority, ranking or score of the search results under the corresponding query sequence etc.; preferably, the mapping relationship between the query sequence and the search result also includes the weight or decision threshold of the feature component of the search result under the corresponding query sequence. For example, for a given model prototype, such as a feature vector containing multiple feature components, but the parameters of each feature component have not been calibrated, such as the weight or decision threshold of the feature component, use the query sequence and search The training set of the mapping relationship between the results, through genetic algorithm, neural network, decision tree, support vector machine and other machine learning algorithms, determine the model parameters including the parameters corresponding to each feature component, that is, obtain the ranking model, as described The first ranking model or the second ranking model.

Preferably, this embodiment further includes step s6', wherein, in step s6', the result providing device determines one or more methods used to determine the The ranking submodel of the feature component in the ranking model described above. Those skilled in the art should be able to understand that, except for the difference between the sorting sub-model and the sorting model, the implementation of step s6' is the same or basically similar to step s5', so for the sake of brevity, it will not be described in detail, and it is only included here by reference . Wherein, the second search result training data includes but not limited to query sequence, search result, and the mapping relationship between query sequence and search result, such as the priority, ranking or score of the search result under the corresponding query sequence, etc. ; Preferably, the mapping relationship between the query sequence and the search result also includes the weight or decision threshold of the feature component of the search result under the corresponding query sequence. For example, for a given sub-model prototype, such as a feature vector containing multiple feature components, but the parameters of each feature component have not been calibrated, such as the weight or decision threshold of the feature component, use the query sequence and The training set of the mapping relationship between search results, through machine learning algorithms such as genetic algorithm, neural network, decision tree, support vector machine, etc., determine the model parameters including the parameters corresponding to each feature component, that is, obtain the sub-ranking model.

Preferably, in step s5', the result providing device can also determine the second ranking model through machine learning according to the labeled and sorted training data of the first search results, wherein the second ranking model includes user Behavioral feature components. Specifically, the user behavior characteristic information includes but not limited to the time information of the user's search, the address information confirmed according to the user's IP address, the operation information about the search results generated by the user through clicking, touching, swiping the screen, and the time spent on the page, etc. , the result providing device can use the training data of the first search results marked and sorted, and use machine learning algorithms such as genetic algorithm and neural network to perform machine learning on the user behavior feature information, and determine the feature components under different user behavior feature information The weight or decision threshold of , that is, the second ranking model is obtained. Here, the user behavior feature information may be included in the first search result training data, or may be stored in a third-party device such as a search engine or a search log database connected to the result-providing device via a network, and through such third-party The application program interface (API) provided by the device obtains the user behavior feature information from the third-party devices.

It will be apparent to those skilled in the art that the invention is not limited to the details of the above-described exemplary embodiments, but that the invention can be embodied in other specific forms without departing from the spirit or essential characteristics of the invention. Accordingly, the embodiments should be regarded in all points of view as exemplary and not restrictive, the scope of the invention being defined by the appended claims rather than the foregoing description, and it is therefore intended that the scope of the invention be defined by the appended claims rather than by the foregoing description. All changes within the meaning and range of equivalents of the elements are embraced in the present invention. Any reference sign in a claim should not be construed as limiting the claim concerned. In addition, it is obvious that the word "comprising" does not exclude other elements or steps, and the singular does not exclude the plural. A plurality of units or means stated in the device claims may also be realized by one unit or device through software or hardware. The words first, second, etc. are used to denote names and do not imply any particular order.

Claims (20)

1. one kind by the computer implemented method that is used to provide Search Results, and wherein, this method may further comprise the steps:

A obtains the corresponding initial search result of search sequence with user's input;

B utilizes first order models, in said initial search result, filters out preferred Search Results;

C utilizes second order models, in said preferred Search Results, filters out the optimum search result;

D offers said user with said optimum search result.

2. method according to claim 1, wherein, this method also comprises:

X through the mode of machine learning, confirms order models according to the first Search Results training data through the mark ordering;

Wherein, said order models comprise following at least each:

-said first order models;

-said second order models.

3. method according to claim 2, wherein, this method also comprises:

-according to the second Search Results training data,, confirm one or more ordering submodels that are used for confirming said order models characteristic component through the mode of machine learning through the mark ordering.

4. method according to claim 2, wherein, said step x comprises:

-according to the first Search Results training data,, confirm said second order models through the mode of machine learning through the mark ordering, wherein, said second order models comprises the user behavior characteristic component.

5. according to each described method in the claim 1 to 4, wherein, said step b comprises:

-utilize said first order models, confirm the priority of said initial search result;

-according to the first predetermined amount threshold,, from said initial search result, filter out said preferred Search Results based on the priority of said initial search result, wherein, the quantity of said preferred Search Results satisfies said first amount threshold.

6. method according to claim 5, wherein, this method also comprises:

-confirm rule according to the first predetermined quantity, confirm said first amount threshold;

Wherein, said first quantity confirm rule comprise following at least each:

-based on the quantity of said initial search result, confirm said first amount threshold;

-establish rules really then based on the predetermined amount threshold that is used for definite said optimum search result, confirm said first amount threshold.

7. according to each described method in the claim 1 to 6, wherein, said step c comprises:

-utilize said second order models, confirm the priority of said preferred Search Results;

-according to the second predetermined amount threshold,, from said preferred Search Results, filter out the optimum search result based on the priority of said preferred Search Results, wherein, said optimum search result's quantity satisfies said second amount threshold.

8. method according to claim 7, wherein, this method also comprises:

-confirm rule according to the second predetermined quantity, confirm said second amount threshold;

Wherein, said second quantity confirm rule comprise following at least each:

-based on the terminal attribute of said user's subscriber equipment, confirm said second amount threshold;

-based on the type information of said search sequence, confirm said second amount threshold;

-based on the quantity of said preferred Search Results, confirm said second amount threshold.

9. one kind is used to provide the result of Search Results that equipment is provided, and wherein, this equipment comprises:

Deriving means as a result is used to obtain the corresponding initial search result of search sequence with user's input;

First screening plant is used to utilize first order models, in said initial search result, filters out preferred Search Results;

Second screening plant is used to utilize second order models, in said preferred Search Results, filters out the optimum search result;

Generator is used for said optimum search result is offered said user as a result.

10. result according to claim 9 provides equipment, and wherein, this equipment also comprises:

Model is confirmed device, is used for through the mode of machine learning, confirming order models according to the first Search Results training data through the mark ordering;

Wherein, said order models comprise following at least each:

-said first order models;

-said second order models.

11. result according to claim 10 provides equipment, wherein, this equipment also comprises:

Submodel is confirmed device, is used for according to the second Search Results training data through the mark ordering, through the mode of machine learning, confirms one or more ordering submodels that are used for confirming said order models characteristic component.

12. result according to claim 10 provides equipment, wherein, said model confirms that device is used for:

-according to the first Search Results training data,, confirm said second order models through the mode of machine learning through the mark ordering, wherein, said second order models comprises the user behavior characteristic component.

13. according to each described result in the claim 9 to 12 equipment is provided, wherein, said first screening plant is used for:

-utilize said first order models, confirm the priority of said initial search result;

-according to the first predetermined amount threshold,, from said initial search result, filter out said preferred Search Results based on the priority of said initial search result, wherein, the quantity of said preferred Search Results satisfies said first amount threshold.

14. result according to claim 13 provides equipment, wherein, this equipment also comprises:

First threshold is confirmed device, is used for confirming rule according to the first predetermined quantity, confirms said first amount threshold;

Wherein, said first quantity confirm rule comprise following at least each:

-based on the quantity of said initial search result, confirm said first amount threshold;

-establish rules really then based on the predetermined amount threshold that is used for definite said optimum search result, confirm said first amount threshold.

15. according to each described result in the claim 9 to 14 equipment is provided, wherein, said second screening plant is used for:

-utilize said second order models, confirm the priority of said preferred Search Results;

-according to the second predetermined amount threshold,, from said preferred Search Results, filter out the optimum search result based on the priority of said preferred Search Results, wherein, said optimum search result's quantity satisfies said second amount threshold.

16. result according to claim 15 provides equipment, wherein, this equipment also comprises:

Second threshold value is confirmed device, is used for confirming rule according to the second predetermined quantity, confirms said second amount threshold;

Wherein, said second quantity confirm rule comprise following at least each:

-based on the terminal attribute of said user's subscriber equipment, confirm said second amount threshold;

-based on the type information of said search sequence, confirm said second amount threshold;

-based on the quantity of said preferred Search Results, confirm said second amount threshold.

17. a search engine comprises as each describedly is used to provide the result of Search Results that equipment is provided in the claim 9 to 16.

18. a search engine plug-in unit comprises as each describedly is used to provide the result of Search Results that equipment is provided in the claim 9 to 16.

19. a browser comprises as each describedly is used to provide the result of Search Results that equipment is provided in the claim 9 to 16.

20. a browser plug-in comprises as each describedly is used to provide the result of Search Results that equipment is provided in the claim 9 to 16.

Images