예제 #1
0
  /*
   *  listPostings displays the first n postings for a term in a
   *  field in an index (specified by reader).  Set n to MAX_VALUE
   *  to display all postings.
   */
  static void listPostings(IndexReader reader, String termString, String field, Integer n)
      throws IOException {

    System.out.println("\nPostings:  " + termString + " " + field);

    /*
     *  Prepare to access the index.
     */
    BytesRef termBytes = new BytesRef(termString);
    Term term = new Term(field, termBytes);
    Bits liveDocs = MultiFields.getLiveDocs(reader);

    /*
     *  Lookup the collection term frequency (ctf).
     */
    long df = reader.docFreq(term);
    System.out.println("\tdf:  " + df);

    long ctf = reader.totalTermFreq(term);
    System.out.println("\tctf:  " + ctf);

    if (df < 1) return;

    /*
     *  Lookup the inverted list.
     */
    DocsAndPositionsEnum postings =
        MultiFields.getTermPositionsEnum(reader, liveDocs, field, termBytes);

    /*
     *  Iterate through the first n postings.
     */
    long count = 0;

    while ((count < n) && (postings.nextDoc() != DocIdSetIterator.NO_MORE_DOCS)) {

      System.out.println("\tdocid: " + postings.docID());
      int tf = postings.freq();
      System.out.println("\ttf: " + tf);
      System.out.print("\tPositions: ");

      for (int j = 0; j < tf; j++) {
        int pos = postings.nextPosition();
        System.out.print(pos + " ");
      }

      System.out.println("");

      count++;
    }
    ;

    return;
  }
예제 #2
0
  public static void main(String[] args) throws Exception {
    // the IndexReader object is the main handle that will give you
    // all the documents, terms and inverted index
    IndexReader r = IndexReader.open(FSDirectory.open(new File("index")));

    // You can figure out the number of documents using the maxDoc() function
    System.out.println("The number of documents in this index is: " + r.maxDoc());

    int i = 0;
    // You can find out all the terms that have been indexed using the terms() function
    TermEnum t = r.terms();
    while (t.next()) {
      // Since there are so many terms, let us try printing only term #100000-#100010
      if (i > 100000) System.out.println("[" + i + "] " + t.term().text());
      if (++i > 100010) break;
    }

    // You can create your own query terms by calling the Term constructor, with the field
    // 'contents'
    // In the following example, the query term is 'brute'
    Term te = new Term("contents", "brute");

    // You can also quickly find out the number of documents that have term t
    System.out.println("Number of documents with the word 'brute' is: " + r.docFreq(te));

    // You can use the inverted index to find out all the documents that contain the term 'brute'
    //  by using the termDocs function
    TermDocs td = r.termDocs(te);
    while (td.next()) {
      System.out.println(
          "Document number ["
              + td.doc()
              + "] contains the term 'brute' "
              + td.freq()
              + " time(s).");
    }

    // You can find the URL of the a specific document number using the document() function
    // For example, the URL for document number 14191 is:
    Document d = r.document(14191);
    String url =
        d.getFieldable("path")
            .stringValue(); // the 'path' field of the Document object holds the URL
    System.out.println(url.replace("%%", "/"));

    // -------- Now let us use all of the functions above to make something useful --------
    // The following bit of code is a worked out example of how to get a bunch of documents
    // in response to a query and show them (without ranking them according to TF/IDF)
    Scanner sc = new Scanner(System.in);
    String str = "";
    System.out.print("query> ");
    while (!(str = sc.nextLine()).equals("quit")) {
      String[] terms = str.split("\\s+");
      for (String word : terms) {
        Term term = new Term("contents", word);
        TermDocs tdocs = r.termDocs(term);
        while (tdocs.next()) {
          String d_url =
              r.document(tdocs.doc()).getFieldable("path").stringValue().replace("%%", "/");
          System.out.println("[" + tdocs.doc() + "] " + d_url);
        }
      }
      System.out.print("query> ");
    }
  }