利用JAVA计算TFIDF和Cosine相似度-学习版本

写在前面的话，既然是学习版本，那么就不是一个好用的工程实现版本，整套代码全部使用List进行匹配效率可想而知。

【原文转自】：http://computergodzilla.blogspot.com/2013/07/how-to-calculate-tf-idf-of-document.html，修改了其中一些bug。

P.S：如果不是被迫需要语言统一，尽量不要使用此工程计算TF-IDF，计算2W条短文本，Matlab实现仅是几秒之间，此Java工程要计算良久。。半个小时？甚至更久，因此此程序作为一个学习版本，并不适用于工程实现。。工程试验版本

For beginners doing a project in text mining aches them a lot by various term like :

• TF-IDF
• COSINE SIMILARITY
• CLUSTERING
• DOCUMENT VECTORS

In my earlier post I showed you guys what is Cosine Similarity. I will not talk about Cosine Similarity in this post but rather I will show a nice little code to calculate Cosine Similarity in java.

Many of you must be familiar with Tf-Idf(Term frequency-Inverse Document Frequency).
I will enlighten them in brief.

Term Frequency:
Suppose for a document “Tf-Idf Brief Introduction” there are overall 60000 words and a word Term-Frequency occurs 60times.
Then , mathematically, its Term Frequency, TF = 60/60000 =0.001.

Inverse Document Frequency:
Suppose one bought Harry-Potter series, all series. Suppose there are 7 series and a word “AbraKaDabra” comes in 2 of the series.
Then, mathematically, its Inverse-Document Frequency , IDF = 1 +
log(7/2) = …….(calculated it guys, don’t be lazy, I am lazy not you
guys.)

And Finally, TFIDF = TF * IDF;

By mathematically I assume you now know its meaning physically.

Document Vector:
There are various ways to calculate document vectors. I am just giving
you an example. Suppose If I calculate all the term’s TF-IDF of a
document A and store them in an array(list, matrix … in any ordered way,
.. you guys are genius you know how to create a vector. ) then I get an
Document Vector of TF-IDF scores of document A.

The class shown below calculates the Term Frequency(TF) and Inverse Document Frequency(IDF).

1. //TfIdf.java
2. package com.computergodzilla.tfidf;
3. import java.util.List;
4. /**
5. * Class to calculate TfIdf of term.
6. * @author Mubin Shrestha
7. */
8. public class TfIdf {
9. /**
10. * Calculates the tf of term termToCheck
11. * @param totalterms : Array of all the words under processing document
12. * @param termToCheck : term of which tf is to be calculated.
13. * @return tf(term frequency) of term termToCheck
14. */
15. public double tfCalculator(String[] totalterms, String termToCheck) {
16. double count = 0;  //to count the overall occurrence of the term termToCheck
17. for (String s : totalterms) {
18. if (s.equalsIgnoreCase(termToCheck)) {
19. count++;
20. }
21. }
22. return count / totalterms.length;
23. }
24. /**
25. * Calculates idf of term termToCheck
26. * @param allTerms : all the terms of all the documents
27. * @param termToCheck
28. * @return idf(inverse document frequency) score
29. */
30. public double idfCalculator(List<String[]> allTerms, String termToCheck) {
31. double count = 0;
32. for (String[] ss : allTerms) {
33. for (String s : ss) {
34. if (s.equalsIgnoreCase(termToCheck)) {
35. count++;
36. break;
37. }
38. }
39. }
40. return 1 + Math.log(allTerms.size() / count);
41. }
42. }

The class shown below parsed the text documents and split them into
tokens. This class will communicate with TfIdf.java class to calculated
TfIdf. It also calls CosineSimilarity.java class to calculated the
similarity between the passed documents.

1. //DocumentParser.java
2. package com.computergodzilla.tfidf;
3. import java.io.BufferedReader;
4. import java.io.File;
5. import java.io.FileNotFoundException;
6. import java.io.FileReader;
7. import java.io.IOException;
8. import java.util.ArrayList;
9. import java.util.List;
10. /**
11. * Class to read documents
12. *
13. * @author Mubin Shrestha
14. */
15. public class DocumentParser {
16. //This variable will hold all terms of each document in an array.
17. private List<String[]> termsDocsArray = new ArrayList<String[]>();
18. private List<String> allTerms = new ArrayList<String>(); //to hold all terms
19. private List<double[]> tfidfDocsVector = new ArrayList<double[]>();
20. /**
21. * Method to read files and store in array.
22. * @param filePath : source file path
23. * @throws FileNotFoundException
24. * @throws IOException
25. */
26. public void parseFiles(String filePath) throws FileNotFoundException, IOException {
27. File[] allfiles = new File(filePath).listFiles();
28. BufferedReader in = null;
29. for (File f : allfiles) {
30. if (f.getName().endsWith(“.txt”)) {
31. in = new BufferedReader(new FileReader(f));
32. StringBuilder sb = new StringBuilder();
33. String s = null;
34. while ((s = in.readLine()) != null) {
35. sb.append(s);
36. }
37. String[] tokenizedTerms = sb.toString().replaceAll(“[\\W&&[^\\s]]”, “”).split(“\\W+”);   //to get individual terms
38. for (String term : tokenizedTerms) {
39. if (!allTerms.contains(term)) {  //avoid duplicate entry
40. allTerms.add(term);
41. }
42. }
43. termsDocsArray.add(tokenizedTerms);
44. }
45. }
46. }
47. /**
48. * Method to create termVector according to its tfidf score.
49. */
50. public void tfIdfCalculator() {
51. double tf; //term frequency
52. double idf; //inverse document frequency
53. double tfidf; //term requency inverse document frequency
54. for (String[] docTermsArray : termsDocsArray) {
55. double[] tfidfvectors = new double[allTerms.size()];
56. int count = 0;
57. for (String terms : allTerms) {
58. tf = new TfIdf().tfCalculator(docTermsArray, terms);
59. idf = new TfIdf().idfCalculator(termsDocsArray, terms);
60. tfidf = tf * idf;
61. tfidfvectors[count] = tfidf;
62. count++;
63. }
64. tfidfDocsVector.add(tfidfvectors);  //storing document vectors;
65. }
66. }
67. /**
68. * Method to calculate cosine similarity between all the documents.
69. */
70. public void getCosineSimilarity() {
71. for (int i = 0; i < tfidfDocsVector.size(); i++) {
72. for (int j = 0; j < tfidfDocsVector.size(); j++) {
73. System.out.println(“between ” + i + “ and ” + j + “  =  ”
74. + new CosineSimilarity().cosineSimilarity
75. (
76. tfidfDocsVector.get(i),
77. tfidfDocsVector.get(j)
78. )
79. );
80. }
81. }
82. }
83. }

This is the class that calculates Cosine Similarity:

1. //CosineSimilarity.java
2. /*
3. * To change this template, choose Tools | Templates
4. * and open the template in the editor.
5. */
6. package com.computergodzilla.tfidf;
7. /**
8. * Cosine similarity calculator class
9. * @author Mubin Shrestha
10. */
11. public class CosineSimilarity {
12. /**
13. * Method to calculate cosine similarity between two documents.
14. * @param docVector1 : document vector 1 (a)
15. * @param docVector2 : document vector 2 (b)
16. * @return
17. */
18. public double cosineSimilarity(double[] docVector1, double[] docVector2) {
19. double dotProduct = 0.0;
20. double magnitude1 = 0.0;
21. double magnitude2 = 0.0;
22. double cosineSimilarity = 0.0;
23. for (int i = 0; i < docVector1.length; i++) //docVector1 and docVector2 must be of same length
24. {
25. dotProduct += docVector1[i] * docVector2[i];  //a.b
26. magnitude1 += Math.pow(docVector1[i], 2);  //(a^2)
27. magnitude2 += Math.pow(docVector2[i], 2); //(b^2)
28. }
29. magnitude1 = Math.sqrt(magnitude1);//sqrt(a^2)
30. magnitude2 = Math.sqrt(magnitude2);//sqrt(b^2)
31. if (magnitude1 != 0.0 | magnitude2 != 0.0) {
32. cosineSimilarity = dotProduct / (magnitude1 * magnitude2);
33. } else {
34. return 0.0;
35. }
36. return cosineSimilarity;
37. }
38. }

Here’s the main class to run the code:

1. //TfIdfMain.java
2. package com.computergodzilla.tfidf;
3. import java.io.FileNotFoundException;
4. import java.io.IOException;
5. /**
6. *
7. * @author Mubin Shrestha
8. */
9. public class TfIdfMain {
10. /**
11. * Main method
12. * @param args
13. * @throws FileNotFoundException
14. * @throws IOException
15. */
16. public static void main(String args[]) throws FileNotFoundException, IOException
17. {
18. DocumentParser dp = new DocumentParser();
19. dp.parseFiles(“D:\\FolderToCalculateCosineSimilarityOf”); // give the location of source file
20. dp.tfIdfCalculator(); //calculates tfidf
21. dp.getCosineSimilarity(); //calculates cosine similarity
22. }
23. }

You can also download the whole source code from here: Download. （Google Drive）

Overall what I did is, I first calculate the TfIdf matrix of all the
documents and then document vectors of each documents. Then I used those
document vectors to calculate cosine similarity.

You think clarification is not enough. Hit me..
Happy Text-Mining!!

from: http://jacoxu.com/?p=1619