CN107918778A

CN107918778A - A kind of information matching method and relevant apparatus

Info

Publication number: CN107918778A
Application number: CN201610887444.0A
Authority: CN
Inventors: 张昌; 张一昌; 赵争超; 张建伟; 蔡仁贵; 林君; 肖谦; 潘林林
Original assignee: Alibaba Group Holding Ltd
Current assignee: Alibaba Group Holding Ltd
Priority date: 2016-10-11
Filing date: 2016-10-11
Publication date: 2018-04-17
Anticipated expiration: 2036-10-11
Also published as: TW201814556A; CN107918778B; WO2018068648A1

Abstract

The embodiment of the present application provides a kind of information matching method and relevant apparatus, the described method includes：Obtain the first information to be matched and the second information；Label classification tree is obtained, the label classification tree includes at least two layers, and every layer includes at least one label node, and father's label node of each label node is the parent mesh of the label node；The first branch and the second branch are obtained from the label classification tree, the content of the label node and the first information of the lowermost layer of first branch matches, and the label node of the lowermost layer of second branch and the content of second information match；According at least to first branch and second branch in every layer of corresponding matching degree, the matching degree of the first information and second information is calculated.As it can be seen that the matching degree that the embodiment of the present application calculates can reflect the relevance between information, so as to improve matching accuracy rate.

Description

A kind of information matching method and relevant apparatus

Technical field

This application involves field of computer technology, more particularly, to a kind of information matching method and relevant apparatus.

Background technology

Information matches technology is a kind of common computer technology, for obtaining the matching degree between a plurality of information.Information Matching technique is widely used in a variety of Internet scenes, for example, a plurality of being commented for what buyer inputted in websites such as e-commerce Valency information, obtains each bar evaluation information and the matching degree of businessman's subscription information, so as to rapidly by information matches technology Navigate to businessman's evaluation information interested.

A kind of current common information matches mode includes：A plurality of information to be matched is segmented, judges whether phase Same word segmentation result, the matching degree between a plurality of information is calculated according to identical word segmentation result.

Obviously, above- mentioned information matching way can only be judged to whether there is identical word segmentation result between a plurality of information, and It can not reflect between a plurality of information and whether there is relevance.For example, the evaluation information of buyer's input is " service is bad ", and businessman Subscription information is " customer service attitude ", although " service is bad " and " customer service attitude " are serviced in description, has certain association Property, but the matching degree calculated according to above- mentioned information matching way is 0, it is clear that matching accuracy rate is relatively low.

The content of the invention

The technical problem that the application solves is to provide a kind of information matching method and relevant apparatus so that calculated It can reflect the relevance between information with degree, so as to improve matching accuracy rate.

For this reason, the technical solution that the application solves technical problem is：

This application provides a kind of information matching method, including：

Obtain businessman's subscription information and user's evaluation information to be matched；

Label classification tree is obtained, the label classification tree includes at least two layers, and every layer includes at least one label node, often Father's label node of a label node is the parent mesh of the label node；

The first branch and the second branch, the label section of the lowermost layer of first branch are obtained from the label classification tree Point and the content of the user's evaluation information match, and label node and the businessman of the lowermost layer of second branch subscribe to The content of information matches；

According at least to first branch and second branch in every layer of corresponding matching degree, the businessman is calculated The matching degree of subscription information and the user's evaluation information.

Optionally, according at least to first branch and second branch in every layer of corresponding matching degree, calculating The matching degree of businessman's subscription information and the user's evaluation information, including：

According at least to first branch and second branch in every layer of corresponding matching degree, the matching of calculating first Degree；

According at least to first matching degree, the matching of businessman's subscription information and the user's evaluation information is calculated Degree.

Optionally, according at least to first branch and second branch in every layer of corresponding matching degree, calculating First matching degree, including：

According at least to first branch and second branch in every layer of corresponding matching degree, and every layer of power Weight values, calculate the first matching degree.

Optionally, the method further includes：

Obtain the statistical model after training；

The affection index of the user's evaluation information is calculated according to the statistical model；

Calculate the affection index of the user's evaluation information and the degree of approximation of target affection index；

According at least to first branch and second branch in every layer of corresponding matching degree, the user is calculated The matching degree of evaluation information and businessman's subscription information, including：

According at least to first branch and second branch in every layer of corresponding matching degree and described approximate Degree, calculates the matching degree of the user's evaluation information and businessman's subscription information.

Optionally, the method further includes：

The affection index of businessman's subscription information, the emotion of businessman's subscription information are calculated according to the statistical model Index is as the target affection index.

Optionally, according at least to first branch and second branch in every layer of corresponding matching degree and institute The degree of approximation is stated, calculates the matching degree of the user's evaluation information and businessman's subscription information, including：

If the degree of approximation is greater than or equal to first threshold, according at least to first branch with second branch every The corresponding matching degree of layer calculates the matching degree of the user's evaluation information and businessman's subscription information；

If the degree of approximation is less than the first threshold, the matching of the user's evaluation information and businessman's subscription information Spend for 0.

Optionally, the statistical model after training is obtained, including：

Obtain the corresponding classification of the user's evaluation information；

Obtain the statistical model after the corresponding training of the classification.

Optionally, the corresponding classification of the user's evaluation information is obtained, including：

Scene classification tree is obtained, the scene classification tree includes at least two layers, and every layer includes at least one scenario node, often Father's scenario node of a scenario node is the parent mesh of the scenario node；

The scenario node with the user's evaluation information matches is obtained from the scene classification tree, determines the matching The corresponding upper level of scenario node or multistage father's scenario node, using the upper level or multistage father's scenario node as the use The corresponding classification of family evaluation information.

Optionally, the method further includes：

Obtain the term vector of the user's evaluation information and the term vector of businessman's subscription information；

The matching degree of the term vector of the user's evaluation information and the term vector of businessman's subscription information is calculated, as Two matching degrees；

According at least to first branch and second branch in every layer of corresponding matching degree and second described With degree, the matching degree of the user's evaluation information and businessman's subscription information is calculated.

Optionally, the method further includes：

Obtain the matching degree between multiple label nodes in the label classification tree；

According between the multiple label node matching degree carry out machine learning, according to the result of machine learning generation or Person corrects the label classification tree.

Present invention also provides a kind of information matching method, including：

Obtain the statistical model after training；

According at least to the degree of approximation of the affection index and target affection index of the user's evaluation information, the user is calculated The matching degree of evaluation information and businessman's subscription information.

Optionally, the method further includes：

Obtain the initial matching degree of the user's evaluation information and businessman's evaluation information；

According at least to the degree of approximation of the affection index and target affection index of the user's evaluation information, the user is calculated The matching degree of evaluation information and businessman's subscription information, including：

According at least to the degree of approximation and the initial matching degree, calculate the user's evaluation information and the businessman subscribes to The matching degree of information.

Optionally, according at least to the degree of approximation and the initial matching degree, the user's evaluation information and described is calculated The matching degree of businessman's subscription information, including：

If the degree of approximation is greater than or equal to first threshold, the user's evaluation is calculated according at least to the initial matching degree The matching degree of information and businessman's subscription information；

Optionally, the statistical model after training is obtained, including：

Obtain the corresponding classification of the user's evaluation information；

Optionally, the method further includes：

The affection index of businessman's subscription information is calculated according to the statistical model, by the feelings of businessman's subscription information Index is felt as the target affection index.

Present invention also provides a kind of data inputting method, including：

Client obtains user's evaluation information or businessman's subscription information input by user；

The client sends the user's evaluation information or businessman's subscription information to computing unit, described to calculate list Member is used for the matching degree for calculating user's evaluation information and businessman's subscription information.

Obtain the first information to be matched and the second information；

The first branch and the second branch, the label section of the lowermost layer of first branch are obtained from the label classification tree The content of point and the first information matches, and the label node of the lowermost layer of second branch is interior with second information Appearance matches；

According at least to first branch and second branch in every layer of corresponding matching degree, calculating described first The matching degree of information and second information.

Optionally, according at least to first branch and second branch in every layer of corresponding matching degree, calculating The matching degree of the first information and second information, including：

According at least to first matching degree, the matching degree of the first information and second information is calculated.

Optionally, the method further includes：

Obtain the statistical model after training；

The affection index of the first information is calculated according to the statistical model；

Calculate the affection index of the first information and the degree of approximation of target affection index；

According at least to first branch and second branch in every layer of corresponding matching degree, calculating described first The matching degree of information and second information, including：

According at least to first branch and second branch in every layer of corresponding matching degree and described approximate Degree, calculates the matching degree of the first information and second information.

Optionally, the method further includes：

The affection index of second information, the affection index conduct of second information are calculated according to the statistical model The target affection index.

Optionally, according at least to first branch and second branch in every layer of corresponding matching degree and institute The degree of approximation is stated, calculates the matching degree of the first information and second information, including：

If the degree of approximation is greater than or equal to first threshold, according at least to first branch with second branch every The corresponding matching degree of layer calculates the matching degree of the first information and second information；

If the degree of approximation is less than the first threshold, the matching degree of the first information and second information is 0.

Optionally, the statistical model after training is obtained, including：

Obtain the corresponding classification of the first information；

Optionally, the corresponding classification of the first information is obtained, including：

Acquisition and the matched scenario node of the first information, determine the matched field from the scene classification tree The corresponding upper level of scape node or multistage father's scenario node, using the upper level or multistage father's scenario node as the described first letter Cease corresponding classification.

Optionally, the training characteristics of the statistical model after the training include the word segmentation result of input information；

The method further includes：The first information is segmented, obtains the word segmentation result of the first information；

The affection index of the first information is calculated according to the statistical model, including：By the participle of the first information As a result the statistical model is input to, obtains the affection index of the first information.

Optionally, the word segmentation result of the input information is that each two adjacent character in the input information is segmented Obtained word segmentation result；

It is described that the first information is segmented, including：Each two adjacent character in the first information is divided Word.

Optionally, the training characteristics of the statistical model after the training further include the affective characteristics of context；

The method further includes：Extract the affective characteristics of the context of the first information；

The word segmentation result of the first information is input to the statistical model, the emotion for obtaining the first information refers to Number, including：By the word segmentation result of the first information and the affective characteristics of the context of the first information, the system is input to Model is counted, obtains the affection index of the first information.

Optionally, the affective characteristics of the context includes any one of following or multinomial：

The Topic Similarity of the affection index of previous sentence, previous sentence and current sentence, overall emotion above distribution, Yi Jishang The emotion distribution of at least one related sentence in text, the Topic Similarity of described at least one related sentence and current sentence are more than second Threshold value.

Optionally, the statistical model after the training includes the first statistical model and the second statistical model after training, institute Stating the training characteristics of the first statistical model includes the word segmentation result of input information, and the training characteristics of second statistical model include The affective characteristics of context.

Optionally, the statistical model after the training is the maximum entropy model after training.

Optionally, the method further includes：

Obtain the term vector of the first information and the term vector of second information；

The matching degree of the term vector of the first information and the term vector of second information is calculated, as the second matching Degree；

According at least to first branch and second branch in every layer of corresponding matching degree and second described With degree, the matching degree of the first information and second information is calculated.

Optionally, the method further includes：

Present invention also provides a kind of information matches device, including：

Information acquisition unit, for obtaining businessman's subscription information and user's evaluation information to be matched；

Classification tree acquiring unit, for label classification tree, the label classification tree includes at least two layers, and every layer is included at least One label node, father's label node of each label node are the parent mesh of the label node；

Branch acquiring unit, for obtaining the first branch and the second branch, first tree from the label classification tree The label node and the content of the user's evaluation information of the lowermost layer of branch match, the label of the lowermost layer of second branch Node and the content of businessman's subscription information match；

Matching degree computing unit, for corresponding at every layer according at least to first branch and second branch Matching degree, calculates the matching degree of businessman's subscription information and the user's evaluation information.

Optionally, the matching degree computing unit is specifically used for, according at least to first branch and second branch In every layer of corresponding matching degree, the first matching degree is calculated, according at least to first matching degree, the businessman is calculated and subscribes to The matching degree of information and the user's evaluation information.

Optionally, according at least to first branch and second branch in every layer of corresponding matching degree, calculating During the first matching degree, the matching degree computing unit is specifically used for, and exists according at least to first branch and second branch Every layer of corresponding matching degree, and every layer of weighted value, calculate the first matching degree.

Optionally, further include：

Model acquiring unit, for obtaining the statistical model after training；

Affection computation unit, for calculating the affection index of the user's evaluation information according to the statistical model；

Degree of approximation computing unit, the affection index for calculating the user's evaluation information are approximate with target affection index Degree；

The matching degree computing unit is specifically used for, according at least to first branch and second branch at every layer point Not corresponding matching degree and the degree of approximation, calculate the matching degree of the user's evaluation information and businessman's subscription information.

Optionally, the affection computation unit, is additionally operable to calculate businessman's subscription information according to the statistical model Affection index, the affection index of businessman's subscription information is as the target affection index.

Optionally, according at least to first branch and second branch in every layer of corresponding matching degree and institute The degree of approximation is stated, when calculating the matching degree of the user's evaluation information and businessman's subscription information, the matching degree computing unit It is specifically used for：

Optionally, the model acquiring unit is specifically used for, and obtains the corresponding classification of the user's evaluation information, obtains institute State the statistical model after the corresponding training of classification.

Optionally, when obtaining the corresponding classification of the user's evaluation information, the model acquiring unit is specifically used for：

Optionally, further include：Term vector acquiring unit, for the term vector for obtaining the user's evaluation information and the business The term vector of family's subscription information；

Matching degree computing unit, is additionally operable to calculate the term vector of the user's evaluation information and businessman's subscription information The matching degree of term vector, as the second matching degree；

According at least to first branch and second branch in every layer of corresponding matching degree, the user is calculated During the matching degree of evaluation information and businessman's subscription information, matching degree computing unit is specifically used for, according at least to described first Branch and second branch calculate the user's evaluation information in every layer of corresponding matching degree and second matching degree With the matching degree of businessman's subscription information.

Optionally, further include：

Amending unit, for obtaining the matching degree in the label classification tree between multiple label nodes, according to described more Matching degree between a label node carries out machine learning, generates or correct the label classification according to the result of machine learning Tree.

Model acquiring unit, for obtaining the statistical model after training；

Matching degree computing unit, for the affection index according at least to the user's evaluation information and target affection index The degree of approximation, calculates the matching degree of the user's evaluation information and businessman's subscription information.

Optionally, further include：

Matching degree acquiring unit, for obtaining the initial matching of the user's evaluation information and businessman's evaluation information Degree；

According at least to the degree of approximation of the affection index and target affection index of the user's evaluation information, the user is calculated During the matching degree of evaluation information and businessman's subscription information, the matching degree computing unit is specifically used for, according at least to described The degree of approximation and the initial matching degree, calculate the matching degree of the user's evaluation information and businessman's subscription information.

Optionally, according at least to the degree of approximation and the initial matching degree, the user's evaluation information and described is calculated During the matching degree of businessman's subscription information, the matching degree computing unit is specifically used for：

Optionally, model acquiring unit is specifically used for, and obtains the corresponding classification of the user's evaluation information, obtains the class Statistical model after the corresponding training of mesh.

Optionally, affection computation unit is additionally operable to, and the emotion of businessman's subscription information is calculated according to the statistical model Index, using the affection index of businessman's subscription information as the target affection index.

Present invention also provides a kind of client, including：

Information acquisition unit, for obtaining user's evaluation information or businessman's subscription information input by user；

Transmitting element, for sending the user's evaluation information or businessman's subscription information to computing unit, the meter Calculate the matching degree that unit is used to calculate user's evaluation information and businessman's subscription information.

Information acquisition unit, for obtaining the first information to be matched and the second information；

Branch acquiring unit, for obtaining the first branch and the second branch, first tree from the label classification tree The content of the label node and the first information of the lowermost layer of branch matches, the label node of the lowermost layer of second branch Match with the content of second information；

Matching degree computing unit, for corresponding at every layer according at least to first branch and second branch Matching degree, calculates the matching degree of the first information and second information.

Optionally, the matching degree computing unit is specifically used for, according at least to first branch and second branch In every layer of corresponding matching degree, the first matching degree is calculated；According at least to first matching degree, the first information is calculated With the matching degree of second information.

Optionally, further include：

Model acquiring unit, for obtaining the statistical model after training；

Affection computation unit, for calculating the affection index of the first information according to the statistical model；

Degree of approximation computing unit, for calculating the affection index of the first information and the degree of approximation of target affection index；

According at least to first branch and second branch in every layer of corresponding matching degree, calculating described first During the matching degree of information and second information, the matching degree computing unit is specifically used for, according at least to first branch With second branch in every layer of corresponding matching degree and the degree of approximation, the first information and described second are calculated The matching degree of information.

Optionally, affection computation unit is additionally operable to, and the affection index of second information is calculated according to the statistical model, The affection index of second information is as the target affection index.

Optionally, according at least to first branch and second branch in every layer of corresponding matching degree and institute The degree of approximation is stated, when calculating the matching degree of the first information and second information, the matching degree computing unit is specifically used for：

Optionally, model acquiring unit is specifically used for, and obtains the corresponding classification of the first information, obtains the classification pair Statistical model after the training answered.

Optionally, when obtaining the corresponding classification of the first information, model acquiring unit is specifically used for：

Described device further includes：Participle unit, for being segmented to the first information, obtains the participle of the first information As a result；

Affection computation unit is specifically used for, and the word segmentation result of the first information is input to the statistical model, is obtained The affection index of the first information.

When being segmented to the first information, participle unit is specifically used for, adjacent to each two in the first information Character is segmented.

Described device further includes：Emotion extraction unit, the affective characteristics of the context for extracting the first information；

The word segmentation result of the first information is input to the statistical model, obtains the affection index of the first information When, affection computation unit is specifically used for, by the word segmentation result of the first information and the emotion of the context of the first information Feature, is input to the statistical model, obtains the affection index of the first information.

Optionally, it is characterised in that the statistical model after the training is the maximum entropy model after training.

Optionally, further include：Term vector acquiring unit, for the term vector for obtaining the first information and second letter The term vector of breath；

Matching degree computing unit, is additionally operable to calculate the term vector and the term vector of second information of the first information Matching degree, as the second matching degree；

According at least to first branch and second branch in every layer of corresponding matching degree, calculating described first During the matching degree of information and second information, matching degree computing unit is specifically used for, according at least to first branch and institute The second branch is stated in every layer of corresponding matching degree and second matching degree, calculates the first information and second letter The matching degree of breath.

Optionally, further include：Amending unit, for obtaining the matching in the label classification tree between multiple label nodes Degree, machine learning is carried out according to the matching degree between the multiple label node, is generated or is repaiied according to the result of machine learning Just described label classification tree.

According to the above-mentioned technical solution, in the embodiment of the present application when matching the first information and the second information, no longer will Directly matched after the first information and the second information participle, but corresponding first branch of the first information is obtained from label classification tree The second branch corresponding with the second information.Wherein, the content of the label node of the lowermost layer of the first branch and the first information Match, and father's label node of each label node is the parent mesh of the label node in the label classification tree, therefore First branch not only includes the label node to match with the content of the first information, further includes the label node to match Successively parent mesh, similarly, second branch not only includes the label node to match with the content of the second information, also wraps The successively parent mesh of the label node to match is included, therefore, according to first branch and second branch at every layer point The first information and the matching degree of the second information that not corresponding matching degree calculates, can not only reflect the first information and the second letter The matching degree of breath, additionally it is possible to reflect the successively parent purpose matching degree of the first information and the second information, equivalent to reflecting first Relevance between information and the successively parent mesh of the second information, so as to improve matching accuracy rate.

Brief description of the drawings

In order to illustrate more clearly of the technical solution in the embodiment of the present application, make required in being described below to embodiment Attached drawing is briefly described, it should be apparent that, drawings in the following description are only some embodiments of the present application, for For those of ordinary skill in the art, other attached drawings can also be obtained according to these attached drawings.

Fig. 1 is a kind of flow diagram for embodiment of the method that the application provides；

Fig. 2 is a kind of schematic diagram for the label classification tree that the application provides；

Fig. 3 is the flow diagram for another embodiment of the method that the application provides；

Fig. 4 is a kind of schematic diagram for the scene classification tree that the application provides；

Fig. 5 is the flow diagram for another embodiment of the method that the application provides；

Fig. 6 is a kind of structure diagram for device embodiment that the application provides；

Fig. 7 is the structure diagram for another device embodiment that the application provides；

Fig. 8 is the structure diagram for another device embodiment that the application provides；

Fig. 9 is the structure diagram for another device embodiment that the application provides；

Figure 10 is the structure diagram for another device embodiment that the application provides；

Figure 11 is the structure diagram for another device embodiment that the application provides.

Embodiment

Evaluation information refers to the feedback information that user inputs in network platforms such as website, APP (application program).For example, After buyer have purchased commodity on e-commerce website, the logistics that the commodity, businessman can be provided, the service procedure such as service into Row evaluation.Businessman can extract businessman's evaluation information interested and be pushed to businessman by inputting businessman's subscription information.Tool Body process includes：Buyer inputs a plurality of evaluation information, and businessman inputs businessman's subscription information, by businessman's subscription information and evaluation information Segmented respectively, judge that both whether there is identical word segmentation result, according to a plurality of information of identical word segmentation result calculating it Between matching degree.

Obviously, above- mentioned information matching way can only be judged between evaluation information and businessman's subscription information with the presence or absence of identical Word segmentation result, and can not reflect and whether there is relevance between the two, such as can not judge between both parent mesh Relevance.For example, the evaluation information of buyer's input is " service is bad ", and businessman's subscription information is " customer service attitude ", though So the parent mesh of " service is bad " and " customer service attitude " is all service, has certain relevance, but according to above- mentioned information It is 0 with the matching degree that mode calculates, it is clear that matching accuracy rate is relatively low, causes businessman to need to have by extra algorithm acquisition The evaluation information of relevance, causes the waste of system resource.

The embodiment of the present application provides a kind of information matching method and relevant apparatus so that the matching degree calculated can reflect Relevance between information, specifically reflects the relevance between the successively parent mesh of a plurality of information, so that it is accurate to improve matching Rate.

It is in order to make those skilled in the art better understand the technical solutions in the application, real below in conjunction with the application The attached drawing in example is applied, the technical solution in the embodiment of the present application is clearly and completely described, it is clear that described implementation Example is merely a part but not all of the embodiments of the present application.It is common based on the embodiment in the application, this area Technical staff's all other embodiments obtained without creative efforts, should all belong to the application protection Scope.

Referring to Fig. 1, the embodiment of the present application provides a kind of embodiment of the method for information matching method, the institute of the present embodiment The method of stating includes：

S101：Obtain the first information to be matched and the second information.

Wherein, the first information and/or second information can be the information such as word input by user, short sentence.Example Such as, the first information can be the user's evaluation information of buyer's input, and second information can be the businessman of businessman's input Subscription information.

S102：Obtain label classification tree.

The label classification tree in the embodiment of the present application includes at least two layers, and every layer includes at least one label node, Father's label node of each label node is the parent mesh of the label node.

Such as the label classification tree shown in Fig. 2 includes three layers, first layer includes a label node：" service ", i.e., it is described The root node of label classification tree；The second layer includes two label nodes：" pre-sales " and " after sale "；Third layer includes four label sections Point：" customer service attitude ", " response speed ", " returning existing " and " guarantee ".Wherein, the label classification tree is according to successively incremental suitable Sequence, corresponding classification successively refine, that is to say, that father's label node of each label node is the parent mesh of the label node. For example, " pre-sales " is the parent mesh of " customer service attitude ", " service " is the parent mesh of " pre-sales ".

S103：The first branch and the second branch are obtained from the label classification tree.First branch and/or described Two branches include at least one label node.

Wherein, the content of the label node and the first information of the lowermost layer of first branch matches, by institute The father's label node for stating each label node in label classification tree is the parent mesh of the label node.Therefore, if first letter It is not root node to cease matched, then first branch not only includes the label node to match with the content of the first information, also Successively parent mesh including the label node to match.

The acquisition process of first branch can include：By the first information with it is each in the label classification tree Node is matched, and obtains matched label node, by the matched label node and the matched label node successively Father node is as first branch.Wherein, before being matched with the label classification tree, the first information can be carried out Participle, word segmentation result is matched with the label classification tree.

For example, the first information is：" service bad ", obtained after the first information is segmented word segmentation result " service " and " bad ", word segmentation result " service " and " bad " are matched with each node in label classification tree, obtain matched mark Node " service " is signed, since the label node " service " is root node, without father node, then " service " is used as the first branch. In another example the first information is：" customer service attitude is bad ", matched label node " visitor is obtained according to above-mentioned similar mode Take attitude ", by the successively father node of " customer service attitude " and " customer service attitude "：" pre-sales " and " service " is used as the first branch.

Likewise, the label node of the lowermost layer of second branch and the content of second information match.If institute State the second information matches is not root node, then second branch not only includes the label to match with the content of the second information Node, further includes the successively parent mesh of the label node to match.The acquisition process of second branch and the described first letter The acquisition process of breath is similar, can include：Second information is matched with each node in the label classification tree, Matched node is obtained, using the successively father node of the matched node and the matched node as second branch.Its In, before being matched with the label classification tree, second information can be segmented, by word segmentation result and the label Classification tree is matched.

S104：According at least to first branch and second branch in every layer of corresponding matching degree, calculating institute State the matching degree of the first information and second information.

Specifically, this step can include：First branch is with second branch in every layer of corresponding matching Degree, calculates the first matching degree；According at least to first matching degree, businessman's subscription information and user's evaluation letter are calculated The matching degree of breath.Matching that can be directly using first matching degree as the first information and the second information in the embodiment of the present application Degree, can also be according to the first matching degree, and combines the matching degree that other specification calculates the first information and the second information.

Since first branch includes at least one layer of label node, second branch includes at least one layer of label section Point, first branch and the corresponding label node of every layer of second branch are matched, and every layer of acquisition is corresponding Matching degree, and according to every layer of corresponding matching degree calculating first information and the matching degree of second information.

For example, first branch includes successively：" service ", second branch include successively：" service ", " pre-sales ", The matching degree of first layer is 100%, and the matching degree of the second layer is 0, and the first matching degree is calculated according to this two layers matching degree.Example Such as using the 1/2 of the sum of this two layers of matching degree as the first information and the matching degree of second information, in above-mentioned example The matching degree calculated is 50%.In another example first branch includes successively：" service ", " pre-sales ", " customer service attitude ", institute State the second branch includes successively：" service ", " pre-sales ", " response speed ", using the 1/3 of the sum of this three layers of matching degree as described in The matching degree of the first information and second information, the matching degree calculated are 67%.

Wherein when calculating the first matching degree according to every layer of corresponding matching degree, it is also contemplated that every layer of weight Value, for example, the first matching degree Tagsim is：

Wherein, w_iFor i-th layer of weighted value, P_iFor first branch with second branch in i-th layer of corresponding matching Degree, P_iFunction I is equal to 1, P when=100%_iWhen ≠ 100%, function I is equal to 0.Wherein, the weighted value of each layer can be all equal to 1, or can also successively be incremented by, weighted value can be configured and/or be adjusted by way of machine learning.Need what is illustrated That above-mentioned formula is only a kind of optional calculation of the first matching degree, those skilled in the art can to above-mentioned formula into Row extension and deformation, such as P_iFunction I can be equal to other numerical value when=100%, or function I can also meet other During part, it is greater than being equal to 1 during certain numerical value, the embodiment of the present application is not limited this.

According to the above-mentioned technical solution, in the embodiment of the present application when matching the first information and the second information, no longer will Directly matched after the first information and the second information participle, but corresponding first branch of the first information is obtained from label classification tree The second branch corresponding with the second information.Wherein, first branch not only includes the mark to match with the content of the first information Node is signed, further includes the successively parent mesh of the label node to match, similarly, second branch not only includes and second The label node that the content of information matches, further includes the successively parent mesh of the label node to match, therefore, according to described The first information and the matching of the second information that first branch is calculated with second branch in every layer of corresponding matching degree Degree, can not only reflect the matching degree of the first information and the second information, additionally it is possible to reflect the first information and the second information successively Parent purpose matching degree, equivalent to the relevance reflected between the first information and the successively parent mesh of the second information, so as to carry High matching accuracy rate.

As it can be seen that the embodiment of the present application is effectively equivalent to add at least one layer of classification to the first information and the second information Label, the matching degree of the first information and the second information is calculated according to the matching degree of the classification label of respective layer.Therefore, using this Shen Please embodiment can calculate belonging to classification have certain relevance information between matching degree, for example, between synonym Matching degree, belongs to matching degree between a plurality of information of same class purpose etc..

For example, the evaluation information of buyer's input is " service is bad ", and businessman's subscription information is " customer service attitude ", though So " service is bad " and " customer service attitude " are serviced in description, when there is certain relevance, but both directly being matched, Matching degree is 0, and matching accuracy rate is relatively low.And when calculating both matching degrees by the embodiment of the present application, first branch is successively Including：" service ", second branch include successively：" service ", " pre-sales ", the matching degree of first layer is 100%, the second layer Matching degree is 0, and the matching degree finally calculated can be 50%.As it can be seen that the matching degree calculated in the embodiment of the present application can be anti- The relevance between both is reflected, therefore improves matching accuracy rate.

It should be noted that in the embodiment of the present application, in addition to user's evaluation information and businessman's subscription information, described One information and second information can also be the information under other application scene.For example, the first information for user micro- The chat message that inputs in letter group, nail nail group, second information be particular subscription information, such as the subscription that group administrator inputs Word subscribes to phrase etc., to this and is not limited in the embodiment of the present application.Said below by a specific example It is bright.

For the wechat group of a film interest group, label classification tree includes two layers, and first layer includes a label section Point：" film ", the second layer include two label nodes：" comedy " and " action is acute ".Wherein, the label classification tree is according to successively Incremental order, corresponding classification successively refine, that is to say, that father's label node of each label node is the label node Parent mesh.For example, " film " is the parent mesh of " comedy " and " action is acute ".If the subscription word of group administrator input is：" film ", Chat message input by user is：" I likes seeing comedy ", will both directly matching when, matching degree 0, matching accuracy rate compared with It is low.And when calculating both matching degrees by the embodiment of the present application, first branch includes successively：" film ", " comedy ", it is described Second branch includes：" film ", the matching degree finally calculated can be 50%, improve matching accuracy rate.

It should be noted that if the first information and/or second information are matched from the label classification tree A plurality of branch, then can choose a branch, from the tree of second information matches from the matched branch of the first information A branch is chosen in branch, the matching degree between branch two-by-two is calculated, using the highest matching degree calculated as described first The matching degree of information and second information.

Information matches mode described in background technology, due to only judging whether identical word segmentation result, nothing Method calculates the matching degree between synonym, and it is relatively low to further result in matching accuracy rate.In order to solve the problems, such as this, it is also proposed that One kind is based on word embedding (Chinese：Term vector) technology information matches mode, pass through word2vec (one kind processing text This double-deck neutral net) the methods of calculate the term vector of information, according to the Similarity measures matching degree between term vector.Cause This embodiment of the present application can be combined with the first information and the second information when calculating the matching degree of the first information and the second information Term vector between similitude.It is specifically described below.

The method can also include：Obtain the term vector of the first information and the term vector of second information；Meter The matching degree of the term vector of the first information and the term vector of second information is calculated, as the second matching degree；In S104 extremely Less according to first matching degree, i.e., described first branch is with second branch in every layer of corresponding matching degree, and institute The second matching degree is stated, calculates the matching degree of the first information and second information.

During specific implementation, after can the first information be segmented, extract the term vector of each word, by the word of each word to Amount is added and obtains the term vector of the first information, the term vector of the second information can be obtained in a comparable manner, more than calculating The modes such as string similarity calculate the matching degree of the term vector of the first information and the term vector of the second information.Wherein, term vector can be with For the term vector extracted using technologies such as word2vec.

, can be by the when calculating the matching degree of the first information and the second information according to the first matching degree and the second matching degree The sum of one matching degree and the second matching degree can also set corresponding weighted value as final matching degree.For example, first The matching degree sim of information and the second information can be：Sim=λ₁Vecsim+λ₁Tagsim, wherein, Tagsim is the first matching Degree, Vecsim are the second matching degree, λ₁And λ₂For corresponding weighted value, which can be carried out by way of machine learning Set and/or adjust.

Wherein, the principle of word embedding technologies is exactly to substantial amounts of information using machine learning techniques Practise, so that word be represented by corresponding term vector, and it is linguistic context residing for word that term vector, which actually represents, but The matching degree calculated under certain situation according to term vector can there are accuracy rate it is relatively low the problem of.Such as a kind of situation, some words The linguistic context of language is although identical, but semanteme but has bigger difference, therefore term vector can not represent exactly under many circumstances The semanteme of word.For example, the semanteme of " good " and " bad " is on the contrary, still the cosine similarity between term vector is but very high.It is such as another A kind of situation, expressed implication is different under various circumstances for identical word.For example, " very thin " is exactly front when describing mobile phone Word, and be exactly negation words when describing down jackets, and be all identical by the matching degree that this mode of term vector calculates.This Outside, due to being difficult the corresponding implication of numerical value proved in term vector, term vector can not be adjusted in itself to solve The certainly above problem.

To solve the above-mentioned problems, the embodiment of the present application can also calculate the affection index of information according to statistical model, should Affection index can indicate that the information is front word, negation words or neutral words, and be examined when calculating final matching degree Consider affection index.

Specifically, as shown in figure 3, the method for the embodiment of the present application can also include：

S301：Obtain the statistical model after training.

Wherein, the statistical model can train to obtain according to substantial amounts of training data, and each training data marks Corresponding affection index.For example, training data is 200,000 sentences, every sentence all marked corresponding affection index.

Optionally, statistical model can be any mathematical models such as maximum entropy model.Largely tested by inventor It was found that during using maximum entropy model, the affection index for enabling to calculate more is bonded semanteme, so as to improve information matches Accuracy rate.

S302：The affection index of the first information is calculated according to the statistical model.

The first information is input to the statistical model after training, the affection index of the first information can be obtained.Wherein, according to The section that affection index is located at, is able to indicate that the corresponding emotion of the first information is respectively positive, negative or neutral.

S303：Calculate the affection index of the first information and the degree of approximation of target affection index.

In the embodiment of the present application, target affection index can be default affection index, can also be according to the second information It is calculated.For example, calculating the affection index of second information according to the statistical model, the emotion of second information refers to Number is used as the target affection index.Target affection index is able to indicate that target emotion is positive, negative or neutral.

Wherein, the degree of approximation can show as difference either any form such as accounting or can also be according to described The affection index of the first information and the target affection index instruction emotion whether identical calculations, if for example, it is described first letter The affection index of breath and the emotion of target affection index instruction are negative, then it represents that both degrees of approximation are higher.

According at least to first branch and second branch in every layer of corresponding matching degree and institute in S104 The degree of approximation is stated, calculates the matching degree of the first information and second information.

In the present embodiment, when calculating the matching degree of the first information and the second information, it is also contemplated that the emotion of the first information The degree of approximation of index and target affection index, and when the degree of approximation is bigger, that is to say, that the emotion of the first information and target feelings Sense closer to when, the matching degree calculated is higher, on the contrary then lower, so as to solve that linguistic context is identical but semantic difference is very big When caused by matching accuracy rate it is low the problem of.Such as " big " and " small ", since emotion differs greatly, that calculates It is also lower with spending, it is consistent with semanteme, so as to improve matching accuracy rate.

Therefore, therefore, can in this example, it is assumed that businessman concerns the unfavorable ratings information in user's evaluation information Using goal-selling affection index as negative corresponding affection index, if user's evaluation information and target affection index are relatively When, then the matching degree finally calculated is higher, so as to extract the unfavorable ratings information of businessman's care according to this mode.

, can be in the following ways when specifically calculating matching degree：

If the degree of approximation is greater than or equal to first threshold, according at least to first branch with second branch every The corresponding matching degree of layer calculates the matching degree of the first information and second information.Such as the feelings of the first information Sense index and the emotion of target affection index instruction are negative, and sim=Tagsim, wherein sim are the first information and the The matching degree of two information, Tagsim are the first matching degree.

If the degree of approximation is less than the first threshold, the matching degree of the first information and second information is 0.Example The emotion difference that affection index such as the first information is indicated with the target affection index, sim=0.At this time described first The matching degree of information and second information can also be other relatively low numerical value, and the embodiment of the present application does not limit this.

In the embodiment of the present application, for identical word, expressed implication is different under various circumstances, can also set more The corresponding statistical model of a classification, each statistical model can be calculated at such now, the affection index of the first information. Different statistical models trains to obtain according to the different corresponding training datas of scene classification, such as same sentence, not It is different with the affection index marked under scene classification, so that affection index and scene class that different statistical models calculates Mesh corresponds to.

Specifically, obtaining the statistical model after training can include：The corresponding classification of the first information is obtained, obtains institute State the statistical model after the corresponding training of classification.Wherein, the corresponding classification of the first information, also refers to first letter Classification belonging to the evaluation object of breath, for example, buyer have purchased the commodity of clothing on e-commerce website, and inputs User's evaluation information is used for the commodity for evaluating the clothing, i.e. classification corresponding to the user's evaluation information is clothing.

Wherein it is possible to the corresponding classification of the first information is obtained by way of scene classification tree.Specifically, institute is obtained Stating the corresponding classification of the first information includes：Scene classification tree is obtained, the scene classification tree includes at least two layers, and every layer is included extremely A few scenario node, father's scenario node of each scenario node are the parent mesh of the scenario node；From the scene classification tree Middle acquisition and the matched scenario node of the first information, determine the corresponding upper level of the matched scenario node or multistage Father's scenario node, using the upper level or multistage father's scenario node as the corresponding classification of the first information.Wherein, upper level Or multistage father's scenario node also refers to root scenario node, that is, directly acquires root scenario node as corresponding classification.

For example, buyer have purchased skirt on e-commerce website, and it have input user's evaluation information and be used to evaluate The skirt, therefore get matched scenario node from scene classification tree：Skirt, determines corresponding field of the scenario node Scape node：Clothing, obtains the statistical model after the corresponding training of clothing, and the feelings of the first information are calculated using the statistical model Feel index.Therefore, the present embodiment is specifically hand according to " very thin " corresponding scene classification when calculating the affection index of " very thin " Machine or clothing, choose corresponding statistical model, so as to calculate the affection index of " very thin " according to scene classification, improve The accuracy rate of information matches.

Optionally, the training characteristics of the statistical model in the present embodiment include the word segmentation result of input information；

The method further includes：The first information is segmented, obtains the word segmentation result of the first information；According to described Statistical model calculates the affection index of the first information, including：The word segmentation result of the first information is input to the system Model is counted, obtains the affection index of the first information.

Largely test and show by inventor, when being segmented, can be segmented based on bigram patterns, also It is that each two adjacent character in the first information is segmented, obtains the word segmentation result of the first information.Such as：" service is not Word segmentation result well " is " service ", " business is not " and " bad ".Higher information matches can be obtained by carrying out participle based on which Accuracy rate.

In addition to word segmentation result, the training characteristics of statistical model can also include the affective characteristics of context, so as to Enough comprehensive word in itself calculates affection index with contextual information.Specifically, the method further includes：Extract described The affective characteristics of the context of one information；The word segmentation result of the first information is input to the statistical model, is obtained described The affection index of the first information, including：By the word segmentation result of the first information and the emotion of the context of the first information Feature, is input to the statistical model, obtains the affection index of the first information.

Wherein, the affective characteristics of the context includes any one of following or multinomial：

The Topic Similarity of the affection index of previous sentence, previous sentence and current sentence, overall emotion above distribution, Yi Jishang The emotion distribution of at least one related sentence in text, the Topic Similarity of described at least one related sentence and current sentence are more than second Threshold value.Illustrate separately below.The affection index of previous sentence can indicate that the emotion of previous sentence is positive, negative or neutral；Before Whether what the Topic Similarity of one and current sentence can represent previous sentence and the description of current sentence is same or similar theme；Above The distribution of overall emotion also refer to above, emotion is respectively positive, negative and neutral sentence quantity；Related sentence is used In representing to describe the sentence of same or similar theme with current sentence, and the emotion distribution of at least one related sentence above can be with Refer to being described above in the sentence of same or similar theme, the quantity of respectively positive, negative and neutral sentence.

The embodiment of the present application can specifically use the affection index of two statistical models calculating first information.That is, Statistical model after the training includes the first statistical model and the second statistical model after training, first statistical model Training characteristics include the word segmentation result of input information, and the training characteristics of second statistical model include the emotion spy of context Sign.

Below by taking the corresponding scene of e-commerce website as an example, a kind of specific embodiment of description the application offer.

Referring to Fig. 5, the embodiment of the present application provides another embodiment of the method for information matching method, the present embodiment The described method includes：

S501：Obtain the user's evaluation information of buyer's input and businessman's subscription information of businessman's input.Wherein, buyer inputs User's evaluation information be used for evaluate buyer purchase skirt, i.e. evaluation object is skirt.

For example, the user is evaluated as " response speed is slow ", businessman's subscription information is " customer service attitude "

S502：Obtain label classification tree as shown in Figure 2.Wherein it is possible to manually the mode such as addition is real to the application The label classification tree applied in example is modified.

S503：The first branch and the second branch are obtained from the label classification tree.The lowermost layer of first branch Label node and the user's evaluation information matches, specifically include：Service, pre-sales, response speed；Second branch it is minimum The label node of layer is matched with businessman's subscription information, is specifically included：Service, pre-sales, customer service attitude.

S503：According at least to first branch and second branch in every layer of corresponding matching degree, the is calculated One matching degree.

For example, the calculation formula of first matching degree is：

Wherein, w_iFor i-th layer of weighted value, P_iFor first branch with it is described Second branch is in i-th layer of corresponding matching degree, P_iFunction I is equal to 1, P when=100%_iWhen ≠ 100%, function I is equal to 0.

S504：The term vector of user's evaluation information and the term vector of businessman's subscription information are obtained respectively, calculate term vector Matching degree, as the second matching degree.

S505：Obtain scene classification tree as shown in Figure 4.Wherein it is possible to manually the mode such as addition is real to the application The scene classification tree applied in example is modified.

S506：Obtained and the matched scenario node of the evaluation object from scene classification tree：Skirt, determines the scene Corresponding scenario node of node：Clothing.

S507：Obtain the maximum entropy model A and maximum entropy model B after the corresponding training of clothing.Maximum entropy model A's Training characteristics include the word segmentation result based on bigram patterns, and the training characteristics of maximum entropy model B include the emotion of context Feature.

S508：User's evaluation information is segmented based on bigram patterns, word segmentation result is input to maximum entropy model A, obtains the affection index of user's evaluation information.

S509：The affective characteristics of the context of user's evaluation information is extracted, by the affective characteristics and S508 of the context Obtained affection index is input to maximum entropy model B, obtains revised affection index.

Wherein, as shown in table 1, the affective characteristics of the context includes following multinomial：

The affection index (respectively positive, negative or neutral, and corresponding intensity) of previous sentence, previous sentence and current Sentence description whether be identical theme, above emotion is respectively positive, negative and neutral sentence quantity and retouches above In the sentence for stating identical theme, the quantity of respectively positive, negative and neutral sentence.

Table 1

S510：User's evaluation information and business are calculated according to revised affection index, the first matching degree and the second matching degree The matching degree of family's subscription information.

Wherein, target emotion is negative, if the emotion of the revised affection index instruction obtained in S509 is not negative, Then matching degree is 0.

If the emotion of the revised affection index instruction obtained in S509 is negative, matching degree is：

Sim=λ₁Vecsim+λ₁Tagsim

Tagsim is the first matching degree for calculating in S503, and Vecsim is the second matching degree calculated in S504, λ₁ And λ₂For corresponding weighted value.

Referring to Fig. 6, the embodiment of the present application additionally provides another embodiment of information matching method.The institute of the present embodiment The method of stating includes：

S601：Obtain the first information to be matched and the second information.

S602：Obtain the statistical model after training.

S603：The affection index of the first information is calculated according to the statistical model.

S604：According at least to the degree of approximation of affection index and the target affection index of the first information, described the is calculated The matching degree of one information and the second information.

Optionally, the method further includes：Obtain the initial matching degree of the first information and second information；Step S604 includes：According at least to the degree of approximation and the initial matching degree, the first information and second information are calculated Matching degree.

Wherein, the initial matching degree can be first matching degree in above-described embodiment, i.e., described first branch With second branch in every layer of corresponding matching degree.

Optionally, according at least to the degree of approximation and the initial matching degree, the first information and described second are calculated The matching degree of information, including：

If the degree of approximation is greater than or equal to first threshold, the first information is calculated according at least to the initial matching degree With the matching degree of second information；

Optionally, the statistical model after training is obtained, including：

Obtain the corresponding classification of the first information；Obtain the statistical model after the corresponding training of the classification.

Optionally, the method further includes：

The affection index of second information is calculated according to the statistical model, the affection index of second information is made For the target affection index.

The related content of the present embodiment please refers to Fig.1, the associated description in 3,5 illustrated embodiments, and which is not described herein again.

A kind of referring to Fig. 7, embodiment present invention also provides data inputting method.The method bag of the present embodiment Include：

S701：Client obtains the first information or the second information.

S702：The client sends the first information or the second information to computing unit, the computing unit For calculating the matching degree of the first information and the second information.

Wherein, computing unit can use any embodiment of above- mentioned information matching process, calculate the first information and the The matching degree of two information.The related content of the present embodiment please refers to Fig.1, the associated description in 3,5 illustrated embodiments, here no longer Repeat.

Corresponding above method embodiment, present invention also provides corresponding device embodiment, is specifically described below.

Referring to Fig. 8, the embodiment of the present application provides a kind of device embodiment of information matches device.The institute of the present embodiment Stating device includes：

Information acquisition unit 801, for obtaining businessman's subscription information and user's evaluation information to be matched.

Classification tree acquiring unit 802, for label classification tree, the label classification tree includes at least two layers, and every layer includes At least one label node, father's label node of each label node are the parent mesh of the label node.

Branch acquiring unit 803, for obtaining the first branch and the second branch from the label classification tree, described first The label node of the lowermost layer of branch and the content of the user's evaluation information match, the mark of the lowermost layer of second branch Label node and the content of businessman's subscription information match.

Matching degree computing unit 804, for right respectively at every layer with second branch according at least to first branch The matching degree answered, calculates the matching degree of businessman's subscription information and the user's evaluation information.

Optionally, further include：

Model acquiring unit, for obtaining the statistical model after training；

Optionally, further include：

Referring to Fig. 9, the embodiment of the present application provides another device embodiment of information matches device.The present embodiment Described device includes：

Information acquisition unit 901, for obtaining businessman's subscription information and user's evaluation information to be matched；

Model acquiring unit 902, for obtaining the statistical model after training；

Affection computation unit 903, for calculating the affection index of the user's evaluation information according to the statistical model；

Matching degree computing unit 904, refers to for the affection index according at least to the user's evaluation information with target emotion Several degrees of approximation, calculates the matching degree of the user's evaluation information and businessman's subscription information.

Optionally, further include：

Referring to Fig. 10, the embodiment of the present application provides a kind of device embodiment of client.The dress of the present embodiment Put including：

Information acquisition unit 1001, for obtaining user's evaluation information or businessman's subscription information input by user；

Transmitting element 1002, for sending the user's evaluation information or businessman's subscription information to computing unit, institute State the matching degree that computing unit is used to calculate user's evaluation information and businessman's subscription information.

1 is please referred to Fig.1, the embodiment of the present application provides another device embodiment of information matches device.The present embodiment Described device include：

Information acquisition unit 1101, for obtaining the first information to be matched and the second information；

Classification tree acquiring unit 1102, for label classification tree, the label classification tree includes at least two layers, and every layer includes At least one label node, father's label node of each label node are the parent mesh of the label node；

Branch acquiring unit 1103, for obtaining the first branch and the second branch from the label classification tree, described The label node of the lowermost layer of one branch and the content of the first information match, the label of the lowermost layer of second branch Node and the content of second information match；

Matching degree computing unit 1104, for right respectively at every layer with second branch according at least to first branch The matching degree answered, calculates the matching degree of the first information and second information.

Optionally, further include：

Model acquiring unit, for obtaining the statistical model after training；

It is apparent to those skilled in the art that for convenience and simplicity of description, the system of foregoing description, The specific work process of device and unit, may be referred to the corresponding process in preceding method embodiment, details are not described herein.

In several embodiments provided herein, it should be understood that disclosed system, apparatus and method can be with Realize by another way.For example, device embodiment described above is only schematical, for example, the unit Division, is only a kind of division of logic function, can there is other dividing mode, such as multiple units or component when actually realizing Another system can be combined or be desirably integrated into, or some features can be ignored, or do not perform.It is another, it is shown or The mutual coupling, direct-coupling or communication connection discussed can be the indirect coupling by some interfaces, device or unit Close or communicate to connect, can be electrical, machinery or other forms.

The unit illustrated as separating component may or may not be physically separate, be shown as unit The component shown may or may not be physical location, you can with positioned at a place, or can also be distributed to multiple In network unit.Some or all of unit therein can be selected to realize the mesh of this embodiment scheme according to the actual needs 's.

In addition, each functional unit in each embodiment of the application can be integrated in a processing unit, can also That unit is individually physically present, can also two or more units integrate in a unit.Above-mentioned integrated list Member can both be realized in the form of hardware, can also be realized in the form of SFU software functional unit.

If the integrated unit is realized in the form of SFU software functional unit and is used as independent production marketing or use When, it can be stored in a computer read/write memory medium.Based on such understanding, the technical solution of the application is substantially The part to contribute in other words to the prior art or all or part of the technical solution can be in the form of software products Embody, which is stored in a storage medium, including some instructions are used so that a computer Equipment (can be personal computer, server, or network equipment etc.) performs the complete of each embodiment the method for the application Portion or part steps.And foregoing storage medium includes：USB flash disk, mobile hard disk, read-only storage (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disc or CD etc. are various can store journey The medium of sequence code.

The above, above example is only to illustrate the technical solution of the application, rather than its limitations；Although with reference to before Embodiment is stated the application is described in detail, it will be understood by those of ordinary skill in the art that：It still can be to preceding State the technical solution described in each embodiment to modify, or equivalent substitution is carried out to which part technical characteristic；And these Modification is replaced, and the essence of appropriate technical solution is departed from the spirit and scope of each embodiment technical solution of the application.

Claims

A kind of 1. information matching method, it is characterised in that including：

Obtain businessman's subscription information and user's evaluation information to be matched；

Label classification tree is obtained, the label classification tree includes at least two layers, and every layer includes at least one label node, Mei Gebiao The father's label node for signing node is the parent mesh of the label node；

Obtain the first branch and the second branch from the label classification tree, the label node of the lowermost layer of first branch with The content of the user's evaluation information matches, the label node of the lowermost layer of second branch and businessman's subscription information Content match；

According at least to first branch and second branch in every layer of corresponding matching degree, calculate the businessman and subscribe to The matching degree of information and the user's evaluation information.
2. according to the method described in claim 1, it is characterized in that, exist according at least to first branch and second branch Every layer of corresponding matching degree, calculates the matching degree of businessman's subscription information and the user's evaluation information, including：

According at least to first branch and second branch in every layer of corresponding matching degree, the first matching degree of calculating；

According at least to first matching degree, the matching degree of businessman's subscription information and the user's evaluation information is calculated.
3. according to the method described in claim 2, it is characterized in that, exist according at least to first branch and second branch Every layer of corresponding matching degree, calculates the first matching degree, including：

According at least to first branch and second branch in every layer of corresponding matching degree, and every layer of weight Value, calculates the first matching degree.
4. according to the method described in claim 1, it is characterized in that, the method further includes：

Obtain the statistical model after training；

The affection index of the user's evaluation information is calculated according to the statistical model；

Calculate the affection index of the user's evaluation information and the degree of approximation of target affection index；

According at least to first branch and second branch in every layer of corresponding matching degree, the user's evaluation is calculated The matching degree of information and businessman's subscription information, including：

According at least to first branch and second branch in every layer of corresponding matching degree and the degree of approximation, meter Calculate the matching degree of the user's evaluation information and businessman's subscription information.
5. according to the method described in claim 4, it is characterized in that, the method further includes：

The affection index of businessman's subscription information, the affection index of businessman's subscription information are calculated according to the statistical model As the target affection index.
6. according to the method described in claim 4, it is characterized in that, exist according at least to first branch and second branch Every layer of corresponding matching degree and the degree of approximation, calculate of the user's evaluation information and businessman's subscription information With degree, including：

If the degree of approximation is greater than or equal to first threshold, according at least to first branch and second branch at every layer point Not corresponding matching degree calculates the matching degree of the user's evaluation information and businessman's subscription information；

If the degree of approximation is less than the first threshold, the matching degree of the user's evaluation information and businessman's subscription information is 0。
7. according to the method described in claim 4, it is characterized in that, obtain training after statistical model, including：

Obtain the corresponding classification of the user's evaluation information；

Obtain the statistical model after the corresponding training of the classification.
8. method according to claim 7, it is characterised in that the corresponding classification of the user's evaluation information is obtained, including：

Scene classification tree is obtained, the scene classification tree includes at least two layers, and every layer includes at least one scenario node, each field Father's scenario node of scape node is the parent mesh of the scenario node；

The scenario node with the user's evaluation information matches is obtained from the scene classification tree, determines the matched field The corresponding upper level of scape node or multistage father's scenario node, the upper level or multistage father's scenario node are commented as the user The corresponding classification of valency information.
9. according to the method described in claim 1, it is characterized in that, the method further includes：

Obtain the term vector of the user's evaluation information and the term vector of businessman's subscription information；

The matching degree of the term vector of the user's evaluation information and the term vector of businessman's subscription information is calculated, as second With degree；

According at least to first branch and second branch in every layer of corresponding matching degree, the user's evaluation is calculated The matching degree of information and businessman's subscription information, including：

According at least to first branch and second branch in every layer of corresponding matching degree and second matching degree, Calculate the matching degree of the user's evaluation information and businessman's subscription information.
10. according to the method described in claim 1, it is characterized in that, the method further includes：

Obtain the matching degree between multiple label nodes in the label classification tree；

Machine learning is carried out according to the matching degree between the multiple label node, generates or repaiies according to the result of machine learning Just described label classification tree.
A kind of 11. information matching method, it is characterised in that including：

Obtain businessman's subscription information and user's evaluation information to be matched；

Obtain the statistical model after training；

The affection index of the user's evaluation information is calculated according to the statistical model；

According at least to the degree of approximation of the affection index and target affection index of the user's evaluation information, the user's evaluation is calculated The matching degree of information and businessman's subscription information.
12. according to the method for claim 11, it is characterised in that the method further includes：

Obtain the initial matching degree of the user's evaluation information and businessman's evaluation information；

According at least to the degree of approximation of the affection index and target affection index of the user's evaluation information, the user's evaluation is calculated The matching degree of information and businessman's subscription information, including：

According at least to the degree of approximation and the initial matching degree, the user's evaluation information and businessman's subscription information are calculated Matching degree.
13. according to the method for claim 12, it is characterised in that according at least to the degree of approximation and the initial matching Degree, calculates the matching degree of the user's evaluation information and businessman's subscription information, including：

If the degree of approximation is greater than or equal to first threshold, the user's evaluation information is calculated according at least to the initial matching degree With the matching degree of businessman's subscription information；

If the degree of approximation is less than the first threshold, the matching degree of the user's evaluation information and businessman's subscription information is 0。
14. according to the method for claim 11, it is characterised in that the statistical model after training is obtained, including：

Obtain the corresponding classification of the user's evaluation information；

Obtain the statistical model after the corresponding training of the classification.
15. according to the method for claim 14, it is characterised in that obtain the corresponding classification of the user's evaluation information, wrap Include：

Scene classification tree is obtained, the scene classification tree includes at least two layers, and every layer includes at least one scenario node, each field Father's scenario node of scape node is the parent mesh of the scenario node；

The scenario node with the user's evaluation information matches is obtained from the scene classification tree, determines the matched field The corresponding upper level of scape node or multistage father's scenario node, the upper level or multistage father's scenario node are commented as the user The corresponding classification of valency information.
16. according to the method for claim 11, it is characterised in that the method further includes：

The affection index of businessman's subscription information is calculated according to the statistical model, the emotion of businessman's subscription information is referred to Number is used as the target affection index.
A kind of 17. data inputting method, it is characterised in that including：

Client obtains user's evaluation information or businessman's subscription information input by user；

The client sends the user's evaluation information or businessman's subscription information to computing unit, and the computing unit is used In the matching degree for calculating user's evaluation information and businessman's subscription information.
A kind of 18. information matching method, it is characterised in that including：

Obtain the first information to be matched and the second information；

Label classification tree is obtained, the label classification tree includes at least two layers, and every layer includes at least one label node, Mei Gebiao The father's label node for signing node is the parent mesh of the label node；

Obtain the first branch and the second branch from the label classification tree, the label node of the lowermost layer of first branch with The content of the first information matches, the label node of the lowermost layer of second branch and the content phase of second information Matching；

According at least to first branch and second branch in every layer of corresponding matching degree, the first information is calculated With the matching degree of second information.
19. according to the method for claim 18, it is characterised in that according at least to first branch and second branch In every layer of corresponding matching degree, the matching degree of the first information and second information is calculated, including：

According at least to first branch and second branch in every layer of corresponding matching degree, the first matching degree of calculating；

According at least to first matching degree, the matching degree of the first information and second information is calculated.
20. according to the method for claim 19, it is characterised in that according at least to first branch and second branch In every layer of corresponding matching degree, the first matching degree is calculated, including：

According at least to first branch and second branch in every layer of corresponding matching degree, and every layer of weight Value, calculates the first matching degree.
21. according to the method for claim 18, it is characterised in that the method further includes：

Obtain the statistical model after training；

The affection index of the first information is calculated according to the statistical model；

Calculate the affection index of the first information and the degree of approximation of target affection index；

According at least to first branch and second branch in every layer of corresponding matching degree, the first information is calculated With the matching degree of second information, including：

According at least to first branch and second branch in every layer of corresponding matching degree and the degree of approximation, meter Calculate the matching degree of the first information and second information.
22. according to the method for claim 21, it is characterised in that the method further includes：

The affection index of second information is calculated according to the statistical model, described in the affection index of second information is used as Target affection index.
23. according to the method for claim 21, it is characterised in that according at least to first branch and second branch In every layer of corresponding matching degree and the degree of approximation, the matching degree of the first information and second information is calculated, Including：

If the degree of approximation is greater than or equal to first threshold, according at least to first branch and second branch at every layer point Not corresponding matching degree calculates the matching degree of the first information and second information；

If the degree of approximation is less than the first threshold, the matching degree of the first information and second information is 0.
24. according to the method for claim 21, it is characterised in that the statistical model after training is obtained, including：

Obtain the corresponding classification of the first information；

Obtain the statistical model after the corresponding training of the classification.
25. according to claim 24 the method, it is characterised in that the corresponding classification of the first information is obtained, including：

Scene classification tree is obtained, the scene classification tree includes at least two layers, and every layer includes at least one scenario node, each field Father's scenario node of scape node is the parent mesh of the scenario node；

Acquisition and the matched scenario node of the first information, determine the matched scene section from the scene classification tree The corresponding upper level of point or multistage father's scenario node, using the upper level or multistage father's scenario node as the first information pair The classification answered.
26. according to the method for claim 21, it is characterised in that the training characteristics of the statistical model after the training include Input the word segmentation result of information；

The method further includes：The first information is segmented, obtains the word segmentation result of the first information；

The affection index of the first information is calculated according to the statistical model, including：By the word segmentation result of the first information The statistical model is input to, obtains the affection index of the first information.
27. according to the method for claim 26, it is characterised in that the word segmentation result of the input information is to the input Each two adjacent character carries out segmenting obtained word segmentation result in information；

It is described that the first information is segmented, including：Each two adjacent character in the first information is segmented.
28. according to the method for claim 26, it is characterised in that the training characteristics of the statistical model after the training are also wrapped Include the affective characteristics of context；

The method further includes：Extract the affective characteristics of the context of the first information；

The word segmentation result of the first information is input to the statistical model, obtains the affection index of the first information, is wrapped Include：By the word segmentation result of the first information and the affective characteristics of the context of the first information, the statistics mould is input to Type, obtains the affection index of the first information.
29. according to the method for claim 28, it is characterised in that the affective characteristics of the context includes any one of following It is or multinomial：

The Topic Similarity of the affection index of previous sentence, previous sentence and current sentence, overall emotion above distribution and above At least one related sentence emotion distribution, the Topic Similarity of described at least one related sentence and current sentence is more than the second threshold Value.
30. according to the method for claim 28, it is characterised in that statistical model after the training includes the after training One statistical model and the second statistical model, the training characteristics of first statistical model include the word segmentation result of input information, institute Stating the training characteristics of the second statistical model includes the affective characteristics of context.
31. according to claim 21 to 30 any one of them method, it is characterised in that the statistical model after the training is instruction Maximum entropy model after white silk.
32. according to the method for claim 18, it is characterised in that the method further includes：

Obtain the term vector of the first information and the term vector of second information；

The matching degree of the term vector of the first information and the term vector of second information is calculated, as the second matching degree；

According at least to first branch and second branch in every layer of corresponding matching degree, the first information is calculated With the matching degree of second information, including：

According at least to first branch and second branch in every layer of corresponding matching degree and second matching degree, Calculate the matching degree of the first information and second information.
33. according to the method for claim 18, it is characterised in that the method further includes：

Obtain the matching degree between multiple label nodes in the label classification tree；

Machine learning is carried out according to the matching degree between the multiple label node, generates or repaiies according to the result of machine learning Just described label classification tree.
A kind of 34. information matches device, it is characterised in that including：

Information acquisition unit, for obtaining businessman's subscription information and user's evaluation information to be matched；

Classification tree acquiring unit, for label classification tree, the label classification tree includes at least two layers, and every layer including at least one Label node, father's label node of each label node are the parent mesh of the label node；

Branch acquiring unit, for obtaining the first branch and the second branch from the label classification tree, first branch The label node of lowermost layer and the content of the user's evaluation information match, the label node of the lowermost layer of second branch Match with the content of businessman's subscription information；

Matching degree computing unit, for according at least to first branch and second branch in every layer of corresponding matching Degree, calculates the matching degree of businessman's subscription information and the user's evaluation information.
A kind of 35. information matches device, it is characterised in that including：

Information acquisition unit, for obtaining businessman's subscription information and user's evaluation information to be matched；

Model acquiring unit, for obtaining the statistical model after training；

Affection computation unit, for calculating the affection index of the user's evaluation information according to the statistical model；

Matching degree computing unit, it is approximate with target affection index for the affection index according at least to the user's evaluation information Degree, calculates the matching degree of the user's evaluation information and businessman's subscription information.
A kind of 36. client, it is characterised in that including：

Information acquisition unit, for obtaining user's evaluation information or businessman's subscription information input by user；

Transmitting element, it is described to calculate list for sending the user's evaluation information or businessman's subscription information to computing unit Member is used for the matching degree for calculating user's evaluation information and businessman's subscription information.
A kind of 37. information matches device, it is characterised in that including：

Information acquisition unit, for obtaining the first information to be matched and the second information；

Classification tree acquiring unit, for label classification tree, the label classification tree includes at least two layers, and every layer including at least one Label node, father's label node of each label node are the parent mesh of the label node；

Branch acquiring unit, for obtaining the first branch and the second branch from the label classification tree, first branch The label node of lowermost layer and the content of the first information match, the label node of the lowermost layer of second branch and institute The content for stating the second information matches；

Computing unit, for according at least to first branch and second branch in every layer of corresponding matching degree, meter Calculate the matching degree of the first information and second information.