用开源Carrot2的后缀树算法做Web文本聚类
采用基于Java的开源搜索结果聚合引擎,Carrot2 2.0 中的后缀树算法
Carrot2 可以自动的把搜索结果归类到相应的语义类别中,这个功能是通过Carrot2一个现成的组件完成的,除此之外Carrot2 还包括了很多其他的搜索结果聚合聚类算法。
因为没有做中文分词,也没有中文的Stopword,所以我们用英文测试,实现代码
1SnippetTokenizer snippetTokenizer = new SnippetTokenizer();
2 List<DocReference> documentReferences = new ArrayList<DocReference>();
3 List<TokenizedDocument> documents = new ArrayList<TokenizedDocument>();
4 TokenizedDocument doc = null;
5 DocReference documentReference = null;
6
7 //从搜索引擎google获取100篇数据
8 {
9 String url = "http://www.google.com/search?as_q=phone&num=100&hl=en&newwindow=1&btnG=Google+Search&as_epq=&as_oq=&as_eq=&lr=&as_ft=i&as_filetype=&as_qdr=all&as_nlo=&as_nhi=&as_occt=any&as_dt=i&as_sitesearch=&as_rights=&safe=images";
10 byte[] pageHtml = HttpUtil.getPage(url);
11 if(pageHtml == null ) return ;
12 try {
13 String strHtml = new String(pageHtml, "utf-8");
14 String[][] result = StringUtil.splitByReg(strHtml,"<td class=j>(.*?)<br>");
15
16 if(result != null)
17 { for(int i=0;i<result.length;i++)
18 {
19 for(int j=0;j<result[i].length;j++)
20 {
21 doc = snippetTokenizer
22 .tokenize(new RawDocumentSnippet(i+"sen"+j,result[i][j].replaceAll("<[^<>]+>",""), "en"));
23 documentReference = new DocReference(doc);
24 documentReferences.add(documentReference);
25 documents.add(doc);
26 }
27 }
28 }
29 } catch (UnsupportedEncodingException e) {
30 e.printStackTrace();
31 }
32 }
33
34
35 //构建后缀树
36 final STCEngine stcEngine = new STCEngine(documentReferences);
37 stcEngine.createSuffixTree();
38 HashMap<String,String> defaults = new HashMap<String,String>();
39 defaults.put("lsi.threshold.clusterAssignment", "0.150");
40 defaults.put("lsi.threshold.candidateCluster", "0.775");
41 final StcParameters params = StcParameters.fromMap(defaults);
42 stcEngine.createBaseClusters(params);
43 stcEngine.createMergedClusters(params);
44
45 final List clusters = stcEngine.getClusters();
46 int max = params.getMaxClusters();
47
48 // Convert STC's clusters to the format required by local interfaces.
49 final List rawClusters = new ArrayList();
50 for (Iterator i = clusters.iterator(); i.hasNext() && (max > 0); max--)
51 {
52 final MergedCluster b = (MergedCluster) i.next();
53 final RawClusterBase rawCluster = new RawClusterBase();
54
55 int maxPhr = 3; // TODO: This should be a configuration parameter moved to STCEngine perhaps.
56 final List phrases = b.getDescriptionPhrases();
57 for (Iterator j = phrases.iterator(); j.hasNext() && (maxPhr > 0); maxPhr--)
58 {
59 Phrase p = (Phrase) j.next();
60 rawCluster.addLabel(p.userFriendlyTerms().trim());
61 }
62
63 for (Iterator j = b.getDocuments().iterator(); j.hasNext();)
64 {
65 final int docIndex = ((Integer) j.next()).intValue();
66 final TokenizedDocument tokenizedDoc = (TokenizedDocument) documents.get(docIndex);
67 final RawDocument rawDoc = (RawDocument) tokenizedDoc.getProperty(TokenizedDocument.PROPERTY_RAW_DOCUMENT);
68 rawCluster.addDocument(rawDoc);
69 }
70
71 rawClusters.add(rawCluster);
72 }
73
74 //得到结果,输出
75 for (Iterator iter = rawClusters.iterator(); iter.hasNext();)
76 {
77 RawCluster cluster = (RawCluster) iter.next();
78 final List phrases = cluster.getClusterDescription();
79 for(int i=0;i<phrases.size();i++)
80 System.out.print("#"+phrases.get(i));
81 System.out.println();
82
83 }
2 List<DocReference> documentReferences = new ArrayList<DocReference>();
3 List<TokenizedDocument> documents = new ArrayList<TokenizedDocument>();
4 TokenizedDocument doc = null;
5 DocReference documentReference = null;
6
7 //从搜索引擎google获取100篇数据
8 {
9 String url = "http://www.google.com/search?as_q=phone&num=100&hl=en&newwindow=1&btnG=Google+Search&as_epq=&as_oq=&as_eq=&lr=&as_ft=i&as_filetype=&as_qdr=all&as_nlo=&as_nhi=&as_occt=any&as_dt=i&as_sitesearch=&as_rights=&safe=images";
10 byte[] pageHtml = HttpUtil.getPage(url);
11 if(pageHtml == null ) return ;
12 try {
13 String strHtml = new String(pageHtml, "utf-8");
14 String[][] result = StringUtil.splitByReg(strHtml,"<td class=j>(.*?)<br>");
15
16 if(result != null)
17 { for(int i=0;i<result.length;i++)
18 {
19 for(int j=0;j<result[i].length;j++)
20 {
21 doc = snippetTokenizer
22 .tokenize(new RawDocumentSnippet(i+"sen"+j,result[i][j].replaceAll("<[^<>]+>",""), "en"));
23 documentReference = new DocReference(doc);
24 documentReferences.add(documentReference);
25 documents.add(doc);
26 }
27 }
28 }
29 } catch (UnsupportedEncodingException e) {
30 e.printStackTrace();
31 }
32 }
33
34
35 //构建后缀树
36 final STCEngine stcEngine = new STCEngine(documentReferences);
37 stcEngine.createSuffixTree();
38 HashMap<String,String> defaults = new HashMap<String,String>();
39 defaults.put("lsi.threshold.clusterAssignment", "0.150");
40 defaults.put("lsi.threshold.candidateCluster", "0.775");
41 final StcParameters params = StcParameters.fromMap(defaults);
42 stcEngine.createBaseClusters(params);
43 stcEngine.createMergedClusters(params);
44
45 final List clusters = stcEngine.getClusters();
46 int max = params.getMaxClusters();
47
48 // Convert STC's clusters to the format required by local interfaces.
49 final List rawClusters = new ArrayList();
50 for (Iterator i = clusters.iterator(); i.hasNext() && (max > 0); max--)
51 {
52 final MergedCluster b = (MergedCluster) i.next();
53 final RawClusterBase rawCluster = new RawClusterBase();
54
55 int maxPhr = 3; // TODO: This should be a configuration parameter moved to STCEngine perhaps.
56 final List phrases = b.getDescriptionPhrases();
57 for (Iterator j = phrases.iterator(); j.hasNext() && (maxPhr > 0); maxPhr--)
58 {
59 Phrase p = (Phrase) j.next();
60 rawCluster.addLabel(p.userFriendlyTerms().trim());
61 }
62
63 for (Iterator j = b.getDocuments().iterator(); j.hasNext();)
64 {
65 final int docIndex = ((Integer) j.next()).intValue();
66 final TokenizedDocument tokenizedDoc = (TokenizedDocument) documents.get(docIndex);
67 final RawDocument rawDoc = (RawDocument) tokenizedDoc.getProperty(TokenizedDocument.PROPERTY_RAW_DOCUMENT);
68 rawCluster.addDocument(rawDoc);
69 }
70
71 rawClusters.add(rawCluster);
72 }
73
74 //得到结果,输出
75 for (Iterator iter = rawClusters.iterator(); iter.hasNext();)
76 {
77 RawCluster cluster = (RawCluster) iter.next();
78 final List phrases = cluster.getClusterDescription();
79 for(int i=0;i<phrases.size();i++)
80 System.out.print("#"+phrases.get(i));
81 System.out.println();
82
83 }
下面是输出聚类phone的结果,还不错
#phone
#Phone Number
#yellow pages
#mobile phone
#cell phone
#Phone Book
#area code
#Business
#services
#Wireless
#people
#directory
#telephone
#address
#online