CN102968447A - SEO (search engine optimization) keyword competition level computing method based on decision tree algorithm - Google Patents

Abstract

The invention discloses an SEO (search engine optimization) keyword competition level computing method based on a decision tree algorithm. The SEO keyword competition level computing method based on the decision tree algorithm comprises the following steps of: firstly, selecting factors P1, P2, P3, P4, P5 and P6 which affect the keyword competition level; then, according to the rule that a keyword, P1, P2, P3, P4, P5, P6 and C are one piece of optimized data record, tidying historical optimized data; generalizing the corresponding attribute; forming a training dataset; taking P1-P6 as a non-categorical attribute and C as a categorical attribute; establishing a corresponding decision tree with a C4.5 algorithm; finally, introducing SEO keyword data to be decided into an upper decision tree; and computing a corresponding analysis result. According to the SEO keyword competition level computing method based on the decision tree algorithm, which is disclosed by the invention, the keyword can be quickly and accurately subjected to quantitative analysis so that a suggestion on optimization is provided for SEO optimization personnel, and the working efficiency of the SEO optimization personnel is improved.

Description

SEO keyword degree of contention computing method based on decision Tree algorithms

Technical field

The present invention relates to the keyword degree of contention computing method in a kind of SEO field, particularly based on the SEO keyword degree of contention computing method of decision Tree algorithms.

Background technology

Keyword is the word that the viewer inputs when searching information in search engine, these keywords are at SEO(Search Engine Optimization, search engine optimization) playing the part of important role in, selecting rational keyword can cater to the user and specifically search for target.Only have and select correct keyword, website SEO is walked on correct general orientation.Determine which type of keyword has determined the important subsequent steps such as web site contents planning, link construction.But existing SEO keyword computing method substantially all are qualitative descriptions, have provided the criterion of Keyword Selection such as a lot of documents, conclude have following some: 1. keyword can not be too wide in range; 2. too unexpected winner do not wanted in keyword; The search custom that 3. will meet the user; 4. the keyword of geographic position, adjectival and the keyword title that product or service are provided can be merged, forming long-tail keyword competition degree can be stronger etc.Adopt quilitative method to carry out the SEO key word analysis, need to a great extent to rely on SEO personnel's experience to carry out, accuracy is not high.Therefore, how quickly and accurately SEO keyword competition degree to be carried out quantitative analysis, become SEO personnel major issue in the urgent need to address.

Summary of the invention

The purpose of this invention is to provide a kind of SEO keyword degree of contention computing method based on decision Tree algorithms, overcome and manually carried out by virtue of experience that SEO keyword degree of contention counting yield is low, the problem of poor accuracy.

The technical solution adopted in the present invention is: the SEO keyword degree of contention computing method based on decision Tree algorithms may further comprise the steps:

Whether the results page of (1) choose keyword search results quantity P1, using intitle instruction search to obtain is counted P2, Search Results and is existed to pay and promote number of pages P4, keyword occur in the website in P3, the Search Results first page number of times P5, length keywords P6 as the influence factor of keyword degree of contention C in Search Results first page title;

(2) generation of training dataset: be an optimization data record according to " keyword, P1, P2, P3, P4, P5, P6, C ", put historical optimization data in order, and corresponding attribute is generally changed, form training dataset;

(3) make up decision tree according to training dataset: take P1-P6 as non-category attribute, take C as category attribute, adopt the C4.5 algorithm to construct corresponding decision tree, this decision tree is equivalent to series of rules;

(4) the application decision tree carries out the analysis of keyword degree of contention: SEO keyword data that will be to be made a strategic decision is incorporated in the above-mentioned decision tree goes, and calculates corresponding analysis result.

It is as follows that the property value of described step (2) is generally changed method:

Be four intervals with the P1 Attribute Generalization, that is: S1:[0,500,000), S2:[50 ten thousand, 100 ten thousand), S3:[100 ten thousand, 300 ten thousand), S4:[300 ten thousand, 10,000 ten thousand);

Be Three regions with the P2 Attribute Generalization, that is: T1:[0,100,000), T2:[10 ten thousand, 50 ten thousand), T3:[50 ten thousand, 100 ten thousand), T4:[100 ten thousand, 1,000 ten thousand);

Be two intervals with the P4 Attribute Generalization, that is: U1:[0,5), U2:[5,10];

Be Three regions with the P5 Attribute Generalization, that is: V1:[0,3), and V2:[3,6), V3:[6,10];

Be four intervals with the P6 Attribute Generalization, that is: W1:[0,6), and W2:[6,10), W3:[10,20);

Whether exist the popularization P3 that pays to be divided into by Search Results: Y() if existing; N(does not exist);

C is divided into four ranks: " competing very strong ", " competing stronger ", " competing less ", " competing very little ".Wherein " compete very strong " and refer to owing to this keyword dog-eat-dog, can only optimize this keyword is after 35; " compete stronger " and refer to optimize the keyword rank 15 to 35, " competing less " refers to keyword is optimized to 4 to 12, and " competing very little " is front 3 that this keyword can be optimized to search engine retrieving result.

The C4.5 algorithm is a kind of greedy algorithm in the described step (3), and namely an attribute of selection optimum is as the detection attribute of next stage, and detailed process is as follows:

1) take P1-P6 as non-category attribute, take C as category attribute, according to information gain rate formula, calculates the attribute of current information ratio of profit increase maximum;

2) with the root node of this attribute as tree;

3) number according to this property value is divided into corresponding branch with the data in the training table;

4) for each branch, repeat 1)-3) process, until all data are all used up;

5) with 1)-4) attribute that finds in the process links up, and is exactly a decision tree, is " competing very strong " on the leaf node, " competing stronger ", " competing less ", " competing very little " these four class labels.

Described step (4) is further comprising the steps of:

A). record needs the keyword of prediction, obtains the value of its corresponding influence factor P1-P6, and obtaining of P1-P6 can by input keyword in search engine, check that the form of the indices of Search Results is finished;

B). be retrieved with property value corresponding to decision-making tree root in will recording, and compare with the value of decision-making tree root attribute, determine the attribute of lower one deck take-off point according to result relatively;

C). value corresponding with one deck take-off point attribute under the decision tree in will recording extracts, and with the value comparison of this take-off point, determines the attribute of more lower one deck take-off point according to comparative result;

D). repeat c) process, until all property values are all more complete, perhaps arrived the decision tree leaf node; The corresponding numerical value of this decision tree leaf node is exactly testing result.

Beneficial effect of the present invention: the SEO keyword degree of contention computing method based on decision Tree algorithms provided by the invention, can carry out quantitative analysis to keyword fast and accurately, personnel provide Optimizing Suggestions to SEO optimization, improve the work efficiency that SEO optimizes personnel.

Embodiment

The present invention is described in detail below in conjunction with embodiment.

Based on the SEO keyword degree of contention computing method of decision Tree algorithms, at first choose the factor P1 that affects the keyword degree of contention, P2, P3, P4, P5 and P6, then be an optimization data record according to " keyword; P1, P2, P3; P4; P5, P6, C ", put historical optimization data in order, and corresponding attribute generally changed, form training dataset, again take P1-P6 as non-category attribute, C is category attribute, adopt the C4.5 algorithm to construct corresponding decision tree, SEO keyword data that at last will be to be made a strategic decision is introduced in the upper decision tree and is gone, and calculates corresponding analysis result.

The concrete steps of the method are as follows:

Whether the results page of (1) choose keyword search results quantity P1, using intitle instruction search to obtain is counted P2, Search Results and is existed to pay and promote number of pages P4, keyword occur in the website in P3, the Search Results first page number of times P5, length keywords P6 as the influence factor of keyword degree of contention in Search Results first page title.

Below respectively to the analysis of making explanations of these influence factors:

Keyword search results quantity P1: result of page searching all can show the related pages sum that keyword returns.This result is that the search engine process is calculated all pages of thinking relevant with search word, namely participates in all pages of this keyword competition.In general, number of results is more, illustrates that the Internet resources relevant with this keyword are abundanter, and therefore competition is also just fiercer.

The results page of using intitle instruction search to obtain is counted P2: this number of results refers to retrieve the results page number that obtains by search engine instruction " intitle: keyword ".In some cases, comprise the page that occurs keyword on the page but do not have in the page title to occur among the result that the simplex search keyword returns, although these pages also have some correlativitys, but probably just mention keyword at the page accidentally, for keyword optimization, these pages are very not low for the competitive strength of this particular keywords.The result who adopts this mode to retrieve can reflect the race condition of keyword to a certain extent.

Whether whether Search Results exists to pay: referring to the search results pages right side of face has the paying promotion message if being promoted P3.In general advertiser inside has the professional to do keyword research and advertisement putting, and they are inevitable to have done detailed degree of contention analysis and payoff profile, only has the keyword that can tell on and get a profit, and they just can go to throw in advertisement.Therefore, whether Search Results exists to pay is promoted the degree of contention that has reflected to a certain extent keyword, can be used as one of index of weighing the keyword degree of contention.

Number of pages P4 in the website in the Search Results first page: each result who retrieves has a link, and this link can be page or leaf, list page in the website, or the homepage of website.Significance level is successively decreased successively by page or leaf in homepage, list page, the website, therefore, and the degree of contention that how much also can reflect keyword of interior number of pages in the Search Results first page.

The number of times P5 that keyword occurs in Search Results first page title: in some cases, search key only shows in the result for retrieval summary, but do not comprise keyword in the title, these pages may and keyword between correlativity less, so the number of times that occurs in title of search key also can reflect the degree of contention of this keyword.

Length keywords P6: the keyword of user input is longer, and it is clearer and more definite represent searched targets, often causes retrieval quantity fewer, so longer keyword is often easier is optimized to forward position.

(2) generation of training dataset: be an optimization data record according to " keyword, P1, P2, P3, P4, P5, P6, C ", put historical optimization data in order, and corresponding attribute is generally changed, form training dataset.

The Attribute Generalization method is as follows:

Be four intervals with the P1 Attribute Generalization, that is: S1:[0,500,000), S2:[50 ten thousand, 100 ten thousand), S3:[100 ten thousand, 300 ten thousand), S4:[300 ten thousand, 10,000 ten thousand);

Be Three regions with the P2 Attribute Generalization, that is: T1:[0,100,000), T2:[10 ten thousand, 50 ten thousand), T3:[50 ten thousand, 100 ten thousand), T4:[100 ten thousand, 1,000 ten thousand);

Be two intervals with the P4 Attribute Generalization, that is: U1:[0,5), U2:[5,10];

Be Three regions with the P5 Attribute Generalization, that is: V1:[0,3), and V2:[3,6), V3:[6,10];

Be four intervals with the P6 Attribute Generalization, that is: W1:[0,6), and W2:[6,10), W3:[10,20);

Whether exist the popularization P3 that pays to be divided into by Search Results: Y() if existing; N(does not exist);

C is divided into four ranks: " competing very strong ", " competing stronger ", " competing less ", " competing very little ".Wherein " compete very strong " and refer to owing to this keyword dog-eat-dog, can only optimize this keyword is after 35; " compete stronger " and refer to optimize the keyword rank 15 to 35, " competing less " refers to keyword is optimized to 4 to 12, and " competing very little " is front 3 that this keyword can be optimized to search engine retrieving result.

(3) make up decision tree according to training dataset: take P1-P6 as non-category attribute, take C as category attribute, adopt the C4.5 algorithm to construct corresponding decision tree, this decision tree is equivalent to series of rules.

The C4.5 algorithm is a kind of greedy algorithm in this step, and namely an attribute of selection optimum is as the detection attribute of next stage, and detailed process is as follows:

1) take P1-P6 as non-category attribute, take C as category attribute, according to information gain rate formula, calculates the attribute of current information ratio of profit increase maximum;

2) with the root node of this attribute as tree, the information gain rate that for example calculates P2 is maximum, and then the root node of this decision tree is P2;

3) number according to this property value is divided into corresponding branch with the data in the training table, puts into first subtree branch of root node such as P2 at the record of 100,000-500,000 scopes;

4) for each branch, repeat 1)-3) process, until all data are all used up;

5) with 1)-4) attribute that finds in the process links up, and is exactly a decision tree, is " competing very strong " on the leaf node, " competing stronger ", " competing less ", " competing very little " these four classifications.

The C4.5 algorithm utilizes the information gain rate to come the selection sort attribute, by recursive operation structure decision tree branches.

1. establishing S is training set, is the set of s data sample, and category attribute has m different value Ci, and si is the sample number among the class Ci, and pi is the probability that arbitrary sample belongs to Ci, and estimates with si/s.The expectation information of arbitrary sample classification:

$I(s_1,s_2,...,s_m)=-\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{p}_i{\log }_2\left(p_i\right)$

2. be divided into the entropy of subset by non-category attribute A:

$E\left(A\right)=\underset{i=1}{\overset{v}{\mathrm{\Σ}}}(s_{1j}+...+s_{\mathrm{mj}})/s\×I(s_{1j}+...+s_{\mathrm{mj}})$

Wherein, non-category attribute A have v different value a1, a2 ..., av }.Utilize A with S be divided into the v subset S1, S2 ..., Sv }, wherein Sj comprises the sample that has value aj among the S at A.Sij is the sample number of class Ci among the subset Sj.

3. the information gain of attribute A is:

Gain(A)＝I(s ₁,+s ₂+...+s _m)-E(A)

4. the division information of introducing attribute in the C4.5 algorithm is corrected information gain, and the division information definition is as follows:

$\mathrm{SplitInformation}(A,S)=-\underset{i=1}{\overset{c}{\mathrm{\Σ}}}\frac{\left|S_i\right|}{\left|S\right|}{\log }_2\frac{\left|S_i\right|}{\left|S\right|}$

Wherein, S ₁That attribute A is cut apart S and c sample set forming to Sc.Division information is that S is about the entropy of each value of attribute A.

5. the information gain ratio is:

$\mathrm{GainRatio}(A,S)=\frac{\mathrm{Gain}\left(A\right)}{\mathrm{SplitInformation}(A,S)}$

Attribute P1-P6 is carried out in definition according to information gain ratio formula, and the attribute with the highest information gain rate is elected to be the testing attribute of given S set.Create a root node, and with this attribute flags, to each value establishment branch of attribute, then recurrence is contribute, and finally finishes the structure of decision tree, wherein each node is the attribute that has maximum gain ratio in the attribute.

(4) the application decision tree carries out the analysis of keyword degree of contention: SEO keyword data that will be to be made a strategic decision is incorporated in the above-mentioned decision tree goes, and calculates corresponding analysis result.

This step is further comprising the steps of: a). and record needs the keyword of prediction, obtains the value of its corresponding influence factor P1-P6, and obtaining of P1-P6 can by input keyword in search engine, check that the form of the indices of Search Results is finished;

B). be retrieved with property value corresponding to decision-making tree root in will recording, and compare with the value of decision-making tree root attribute, determine the attribute of lower one deck take-off point according to result relatively;

C). value corresponding with one deck take-off point attribute under the decision tree in will recording extracts, and with the value comparison of this take-off point, determines the attribute of more lower one deck take-off point according to comparative result;

D). repeat c) process, until all property values are all more complete, perhaps arrived the decision tree leaf node; The corresponding numerical value of this decision tree leaf node is exactly testing result.

If certain keyword record is by above-mentioned 1)-4) process, the value of the leaf node that reaches is " competing very strong ", the degree of contention result who then represents for this keyword is " competing very strong ".

Claims (4)

1. based on the SEO keyword degree of contention computing method of decision Tree algorithms, it is characterized in that, may further comprise the steps:

Whether the results page of (1) choose keyword search results quantity P1, using intitle instruction search to obtain is counted P2, Search Results and is existed to pay and promote number of pages P4, keyword occur in the website in P3, the Search Results first page number of times P5, length keywords P6 as the influence factor of keyword degree of contention C in Search Results first page title;

(2) generation of training dataset: be an optimization data record according to " keyword, P1, P2, P3, P4, P5, P6, C ", put historical optimization data in order, and corresponding attribute is generally changed, form training dataset;

(3) make up decision tree according to training dataset: take P1-P6 as non-category attribute, take C as category attribute, adopt the C4.5 algorithm to construct corresponding decision tree, this decision tree is equivalent to series of rules;

(4) the application decision tree carries out the analysis of keyword degree of contention: SEO keyword data that will be to be made a strategic decision is incorporated in the above-mentioned decision tree goes, and calculates corresponding analysis result.

2. SEO keyword degree of contention computing method based on decision Tree algorithms according to claim 1 is characterized in that it is as follows that the property value of described step (2) is generally changed method:

Be four intervals with the P1 Attribute Generalization, that is: S1:[0,500,000), S2:[50 ten thousand, 100 ten thousand), S3:[100 ten thousand, 300 ten thousand), S4:[300 ten thousand, 10,000 ten thousand);

Be Three regions with the P2 Attribute Generalization, that is: T1:[0,100,000), T2:[10 ten thousand, 50 ten thousand), T3:[50 ten thousand, 100 ten thousand), T4:[100 ten thousand, 1,000 ten thousand);

Be two intervals with the P4 Attribute Generalization, that is: U1:[0,5), U2:[5,10];

Be Three regions with the P5 Attribute Generalization, that is: V1:[0,3), and V2:[3,6), V3:[6,10];

Be four intervals with the P6 Attribute Generalization, that is: W1:[0,6), and W2:[6,10), W3:[10,20);

Whether exist the popularization P3 that pays to be divided into by Search Results: Y() if existing; N(does not exist);

C is divided into four ranks: " competing very strong ", " competing stronger ", " competing less ", " competing very little ".Wherein " compete very strong " and refer to owing to this keyword dog-eat-dog, can only optimize this keyword is after 35; " compete stronger " and refer to optimize the keyword rank 15 to 35, " competing less " refers to keyword is optimized to 4 to 12, and " competing very little " is front 3 that this keyword can be optimized to search engine retrieving result.

3. SEO keyword degree of contention computing method based on decision Tree algorithms according to claim 1 and 2, it is characterized in that, the C4.5 algorithm is a kind of greedy algorithm in the described step (3), and namely an attribute of selection optimum is as the detection attribute of next stage, and detailed process is as follows:

1) take P1-P6 as non-category attribute, take C as category attribute, according to information gain rate formula, calculates the attribute of current information ratio of profit increase maximum;

2) with the root node of this attribute as tree;

3) number according to this property value is divided into corresponding branch with the data in the training table;

4) for each branch, repeat 1)-3) process, until all data are all used up;

5) with 1)-4) attribute that finds in the process links up, and is exactly a decision tree, is " competing very strong " on the leaf node, " competing stronger ", " competing less ", " competing very little " these four class labels.

4. SEO keyword degree of contention computing method based on decision Tree algorithms according to claim 3 is characterized in that described step (4) is further comprising the steps of:

A). record needs the keyword of prediction, obtains the value of its corresponding influence factor P1-P6, and obtaining of P1-P6 can by input keyword in search engine, check that the form of the indices of Search Results is finished;

B). be retrieved with property value corresponding to decision-making tree root in will recording, and compare with the value of decision-making tree root attribute, determine the attribute of lower one deck take-off point according to result relatively;

C). value corresponding with one deck take-off point attribute under the decision tree in will recording extracts, and with the value comparison of this take-off point, determines the attribute of more lower one deck take-off point according to comparative result;

D). repeat c) process, until all property values are all more complete, perhaps arrived the decision tree leaf node; The corresponding numerical value of this decision tree leaf node is exactly testing result.