CN106446191A - Logistic regression based multi-feature network popular tag prediction method - Google Patents

Abstract

The invention relates to a Logistic regression based multi-feature network popular tag prediction method. The method includes the following steps: (1) constructing a weighted undirected network Tag network according to question-and-answer website post data; (2) extracting a popular tag set and an unpopular tag set according to tag occurrence frequency; (3) extracting network features of tags, tag proposer attribute characteristics and tag after-propose attribute change characteristics as feature vectors; (4) adopting Logistics multiple regression training, and constructing a tag classification model. Relevance between the tags are taken into consideration, the tags are classified according to multiple features, and precision on prediction of potential popular tags is high. The method is in favor of guiding user to select reasonable tags and beneficial to website designers to provide higher-quality tags.

Description

A kind of multiple features network flow row label Forecasting Methodology based on Logistic recurrence

Technical field

The present invention relates to data mining and field of computer technology, more particularly to a kind of many based on Logistic recurrence Character network popular Tag Estimation method.

Background technology

Web tab (Tag) is a kind of organizational form of internet information content, generally closely related with content by some Keyword composition, it can help people easily to describe and categorised content, also allow for the retrieval of information simultaneously and share.By In the convenience of web tab, Tag Estimation and label recommendations have obtained widely should in recent years in numerous network platforms With such as question and answer website StackExchange, photo sharing website Flickr, and food and drink comment website Yelp.Using suitable Label is either still all extremely important for a user to website.For website, suitable label can help website pair User carries out personalized recommendation, increases viscosity and the website clicking rate of user；For a user, label can help user quick Navigate to needed for oneself, it is to avoid lose time to browse garbage.In label is chosen, how to choose potential popular label is ten Divide crucial step, because popular label often represents the demand of most of user.

The Main Basiss that at present information is entered with row label selection are that information is sent out with the word degree of correlation of label and information Play self attributes of person etc..But there are various drawbacks in such selection, be mainly manifested in：1. have ignored label potential popular become Gesture；2. have ignored the correlation between label and label；3. cold content leads to unexpected winner label so that information can not be effective Search；4. only take into account a few features so that part labels selection tend to unilateral.

Therefore, in order that user preferably chooses to label when releasing news content, choose potential as much as possible Popular label.The multiple features network flow row label Forecasting Methodology that the present invention is returned based on Logistic solves following two substantially to ask Topic：(1) predict the following fashion trend of label；(2) apply substantial amounts of feature that the fashion trend of label is carried out with quantitative portraying.

Content of the invention

In order to overcome existing label selecting system to have ignored correlation, evaluation between the potential fashion trend of label and label The single deficiency of feature, the invention provides a kind of multiple features network flow row label Forecasting Methodology based on Logistic recurrence, Consider not only the correlated characteristic between multiple features and label, the fashion trend of label also can be better anticipated simultaneously.

The technical solution adopted for the present invention to solve the technical problems is as follows：

A kind of multiple features network flow row label Forecasting Methodology based on Logistic recurrence, comprises the steps：

S1：Data prediction：Collect the information content of website and label data, and by site information content temporally ascending order Arrangement, the model that ratio is front α % is considered as the Temporal Data before label network is stablized, and deletes this part of Temporal Data； Before choosing from the remaining data in website, the data of preset ratio is as training data；

S2：Build label Tag network, to the Tag occurring in the same information content so as to formation between any two connects side, All information are traveled through, obtains the label network figure G of Undirected networks that has the right_Tag, the weight of network is the secondary of both common appearance Number；

S3：The frequency descending that each label occurs in model according to it, the Tag taking front β % ratio is as popular Tag set U_PopularTag；

S4：Find non-popular tag set U_UnPopularTag, to each popular label t ∈ U_PopularTag, search for label The time that t occurs for the first time, and centered on this time, search nearest from this time, occur for the first time, be not belonging to simultaneously U_PopularTagLabel as non-streaming row label, the non-streaming row label set U of composition comparison_UnPopularTag；

S5：Sample label set U={ U to training_PopularTag,U_UnPopularTag, extract the network characterization of Tag in it, In the Undirected networks G that has the right_TagOn, extract neighbor node angle value, the neighbor node degree center that sample label occurs connecting for the first time Property；

S6：Sample label set U={ U to training_PopularTag,U_UnPopularTag, the presenter extracting Tag in it belongs to Property feature, specifically include Tag presenter propose during this Tag with issue the information content quantity, the length of the information content；

S7：Sample label set U={ U to training_PopularTag,U_UnPopularTag, the attribute extracting Tag in it changes Feature, after specifically including this Tag proposition, the answer quantity that in 5 days, the corresponding model of this Tag receives；

S8：Using Logistic multiple regression, with set U={ U_PopularTag,U_UnPopularTagIn label feature conduct Training data, trains and builds label classifier model.

Further, in described step S1, the determination mode of α % is, when occurring default the hundred of website whole Tag number of labels Point than when, as the intercept point of α %.Its objective is to guarantee label network be not subject to website to set up at the beginning of staff to website The impact that label debugging causes；

Further, the node angle value of neighbours i in described step S5, is calculated using formula (1)

Wherein, g represents the node total number of network；If node i and j have even side, x_ij=1, otherwise x_ij=0；

Calculate the node degree centrality of neighbours i using formula (2)

Beneficial effects of the present invention are：Consider correlation between label, label is classified, for pre- according to multiple features Survey potential popular label and there is higher precision.Not only improve guiding user and select rational label, be also beneficial to Web Hosting Person provides higher-quality label.

Brief description

Fig. 1 is a kind of flow chart of multiple features labeling method based on Logistic recurrence of the embodiment of the present invention.

Fig. 2 is the label frequency of occurrences schematic diagram of the embodiment of the present invention.

Specific embodiment

With reference to Figure of description, the specific embodiment of the present invention is described in further detail.

See figures.1.and.2, a kind of multiple features network flow row label Forecasting Methodology based on Logistic recurrence, the present invention Enter row label using data disclosed in the Tex.Stackexchange.com official of question and answer website StackExchange subnet station to divide The modeling analysis of class system, the time that each model of original data record occurs, post people ID, the information such as model label.With As a example this patent research label Tag, we extract the time that this label occurs for the first time, label presenter ID, neighbours' tag ID Etc. information.

In the present embodiment, a kind of multiple features labeling method based on Logistic recurrence, it concretely comprises the following steps：

1) build label Tag network：To the model data delivered, do following process：

1.1) travel through model data, obtain all of Tag tag set T_I, I ∈ N, wherein N represent the total quantity of label. Peek amount be N × 20% label as the number of labels needed for web site tags point of safes, its beneficial manner be prevent website from building Vertical part, staff brings noise to the debugging of web site contents to model；

1.2) by model ascending order arrangement sequentially in time, travel through model data again, when the quantity obtaining different labels During for N × 20%, recording now traversed model number is N_{InstablePosts}, the model time of delivering now is considered as website Label stabilization time；

1.3) determineWherein N_PostsFor delivering the total quantity of model；

1.4) build Tag network：The model of α % before removal, reads the model of front 80% data volume in question and answer website data As training data.Wherein, Tag network struction mode is：To the Tag occurring in same model so as to be formed between any two Lian Bian.All information are traveled through, obtains the label network figure G of Undirected networks that has the right_Tag, the weight of network is that both are common to be occurred Number of times；

2) obtain popular tag set U_PopularTag：To the model data delivered, do following process：

2.1) travel through model data, obtain the frequency that each Tag occurs in model；

2.2) according to Tag frequency of occurrences descending, the Tag taking front β % ratio is as popular tag set U_PopularTag, Here, we select β %=5%；

3) obtain non-streaming row label set U_UnPopularTag, concretely comprise the following steps：

3.1) to each label Tag, travel through model, obtain the time of occurrence first of each label；

3.2) to each popular label t ∈ U_PopularTag, (remaining label is not present in popular to search for remaining labels all In label)With the time difference of this label, i.e. remaining poor Δ T of time of occurrence first with this label；

3.3) ascending order arrangement is carried out to this time difference Δ T, take the minimum label t' of Δ T as non-streaming row label, thus shape Become non-streaming row label set U_UnPopularTag

4) extract the network characterization of Tag, concretely comprise the following steps：

4.1) to each label t ∈ { U_PopularTag,U_UnPopularTag, the node degree of neighbours i is calculated using formula (1) Value

Wherein, g represents the node total number of network；If node i and j have even side, x_ij=1, otherwise x_ij=0；

4.2) formula (2) is adopted to calculate the node degree centrality of neighbours i

4.3) normalization neighbor node degree, neighbor node degree centrality, normalization denominator is neighbor node numerical value

5) extract sample Tag presenter's attributive character, concretely comprise the following steps：

5.1) to each sample label t ∈ { U_PopularTag,U_UnPopularTag, when obtaining this label and proposing first, propose No. ID of person, label time of occurrence first；

5.2) by model ascending order arrangement sequentially in time, find out label first before time of occurrence, this presenter ID is total Common enquirement quantity, answer quantity, as Tag presenter's attributive character；

6) extract the attribute variation feature of sample Tag, concretely comprise the following steps：Sample label set U=to training {U_PopularTag,U_UnPopularTag, after this Tag proposes, the answer quantity that in 5 days, this Tag receives altogether；

7) Logistic multiple regression train classification models：By above-mentioned sample label set U={ U_PopularTag, U_UnPopularTag, and the neighbor node angle value of corresponding Tag, neighbor node centrad, Tag presenter put question to quantity, Tag Presenter's answer quantity, Tag propose after this 5 features of answer quantity of receiving of certain time as input, with Logistics Label classifier model is trained and built to multiple regression, as grader,；

It is the present invention as mentioned above in the StackExchange subnet station Tex.Stackexchange.com of question and answer website Labeling embodiment introduce, by build network by way of by label between correlation include feature；By considering to mark Sign neighbors feature, consider that the modes such as label presenter's feature, label temporal evolution feature increased the characteristic of labeling. The whether popular judgement of label is finally given by training pattern, provides directive significance to the label recommendations system constructing of website.

Claims (3)

1. a kind of multiple features network flow row label Forecasting Methodology based on Logistic recurrence is it is characterised in that methods described bag Include following steps：

S1：Data prediction：Collect the information content and the label data of website, and temporally ascending order is arranged by site information content Row, the model that ratio is front α % is considered as the Temporal Data before label network is stablized, and deletes this part of Temporal Data；From Before choosing in the remaining data in website, the data of preset ratio is as training data；

S2：Build label Tag network, to the Tag occurring in the same information content so as to formation between any two connects side；To institute There is information to travel through, obtain the label network figure G of Undirected networks that has the right_Tag, the weight of network is both common number of times occurring；

S3：The frequency descending that each label occurs in model according to it, the Tag taking front β % ratio is as popular label Set U_PopularTag；

S4：Find non-popular tag set U_UnPopularTag, to each popular label t ∈ U_PopularTag, search for label t first The time of secondary appearance, and centered on this time, search nearest from this time, occur for the first time, be not belonging to simultaneously U_PopularTagLabel as non-streaming row label, the non-streaming row label set U of composition comparison_UnPopularTag；

S5：Sample label set U={ U to training_PopularTag,U_UnPopularTag, extract the network characterization of Tag in it, having Power Undirected networks G_TagOn, extract neighbor node angle value, the neighbor node degree centrality that sample label occurs connecting for the first time；

S6：Sample label set U={ U to training_PopularTag,U_UnPopularTag, the presenter's attribute extracting Tag in it is special Levy, specifically include the quantity that Tag presenter proposes the information content to issue during this Tag, the length of the information content；

S7：Sample label set U={ U to training_PopularTag,U_UnPopularTag, extract the attribute variation feature of Tag in it, After specifically including this Tag proposition, the answer quantity that in 5 days, the corresponding model of this Tag receives；

S8：Using Logistic multiple regression, with set U={ U_PopularTag,U_UnPopularTagIn label feature as training Data, trains and builds label classifier model.

2. a kind of multiple features network flow row label Forecasting Methodology based on Logistic recurrence as claimed in claim 1, it is special Levy and be：In described step S1, the determination mode of α % is, when the preset percentage of website whole Tag number of labels Wait, as the intercept point of α %.

3. a kind of multiple features network flow row label Forecasting Methodology based on Logistic recurrence as claimed in claim 1 or 2, its It is characterised by：The node angle value of neighbours i in described step S5, is calculated using formula (1)

$k_i=\underset{j=1}{\overset{g}{\Σ}}{x}_{ij},(i\≠j)---\left(1\right)$

Wherein, g represents the node total number of network；If node i and j have even side, x_ij=1, otherwise x_ij=0；

Calculate the node degree centrality of neighbours i using formula (2)

$C_D\left(i\right)=\frac{k_i}{g-1}---\left(2\right).$