转载自:http://blog.csdn.net/pukuimin1226/article/details/17558247/
1、Nuget Lucene的盘古分析器会自动添加Lucene及分词高亮等引用
2、将自动添加的 Dict 字典文件复制到输出目录
3、添加盘古分词及Lucene帮助类
using Lucene.Net.Analysis; using Lucene.Net.Documents; using Lucene.Net.Index; using Lucene.Net.QueryParsers; using Lucene.Net.Search; using System; using System.Collections.Generic; using System.Linq; using System.Reflection; using System.Web; namespace WebApplication1.Commons { /// <summary> /// 盘古分词在lucene.net中的使用帮助类 /// 调用PanGuLuceneHelper.instance /// </summary> public class PanGuLuceneHelper { private PanGuLuceneHelper() { } #region 单一实例 private static PanGuLuceneHelper _instance = null; /// <summary> /// 单一实例 /// </summary> public static PanGuLuceneHelper instance { get { if (_instance == null) _instance = new PanGuLuceneHelper(); return _instance; } } #endregion #region 分词测试 /// <summary> /// 分词测试 /// </summary> /// <param name="keyword"></param> /// <returns></returns> public string Token(string keyword) { string ret = ""; System.IO.StringReader reader = new System.IO.StringReader(keyword); Lucene.Net.Analysis.TokenStream ts = analyzer.TokenStream(keyword, reader); bool hasNext = ts.IncrementToken(); Lucene.Net.Analysis.Tokenattributes.ITermAttribute ita; while (hasNext) { ita = ts.GetAttribute<Lucene.Net.Analysis.Tokenattributes.ITermAttribute>(); ret += ita.Term + "|"; hasNext = ts.IncrementToken(); } ts.CloneAttributes(); reader.Close(); analyzer.Close(); return ret; } #endregion #region 创建索引 /// <summary> /// 创建索引 /// </summary> /// <param name="datalist"></param> /// <returns></returns> public bool CreateIndex(List<MySearchUnit> datalist) { IndexWriter writer = null; try { writer = new IndexWriter(directory_luce, analyzer, false, IndexWriter.MaxFieldLength.LIMITED);//false表示追加(true表示删除之前的重新写入) } catch { writer = new IndexWriter(directory_luce, analyzer, true, IndexWriter.MaxFieldLength.LIMITED);//false表示追加(true表示删除之前的重新写入) } foreach (MySearchUnit data in datalist) { CreateIndex(writer, data); } writer.Optimize(); writer.Dispose(); return true; } public bool CreateIndex(IndexWriter writer, MySearchUnit data) { try { if (data == null) return false; Document doc = new Document(); Type type = data.GetType();//assembly.GetType("Reflect_test.PurchaseOrderHeadManageModel", true, true); //命名空间名称 + 类名 //创建类的实例 //object obj = Activator.CreateInstance(type, true); //获取公共属性 PropertyInfo[] Propertys = type.GetProperties(); for (int i = 0; i < Propertys.Length; i++) { //Propertys[i].SetValue(Propertys[i], i, null); //设置值 PropertyInfo pi = Propertys[i]; string name = pi.Name; object objval = pi.GetValue(data, null); string value = objval == null ? "" : objval.ToString(); //值 if (name == "id" || name == "flag")//id在写入索引时必是不分词,否则是模糊搜索和删除,会出现混乱 { doc.Add(new Field(name, value, Field.Store.YES, Field.Index.NOT_ANALYZED));//id不分词 } else { doc.Add(new Field(name, value, Field.Store.YES, Field.Index.ANALYZED)); } } writer.AddDocument(doc); } catch (System.IO.FileNotFoundException fnfe) { throw fnfe; } return true; } #endregion #region 在title和content字段中查询数据 /// <summary> /// 在title和content字段中查询数据 /// </summary> /// <param name="keyword"></param> /// <returns></returns> public List<MySearchUnit> Search(string keyword) { string[] fileds = { "title", "content" };//查询字段 //Stopwatch st = new Stopwatch(); //st.Start(); QueryParser parser = null;// new QueryParser(Lucene.Net.Util.Version.LUCENE_30, field, analyzer);//一个字段查询 parser = new MultiFieldQueryParser(version, fileds, analyzer);//多个字段查询 Query query = parser.Parse(keyword); int n = 1000; IndexSearcher searcher = new IndexSearcher(directory_luce, true);//true-表示只读 TopDocs docs = searcher.Search(query, (Filter)null, n); if (docs == null || docs.TotalHits == 0) { return null; } else { List<MySearchUnit> list = new List<MySearchUnit>(); int counter = 1; foreach (ScoreDoc sd in docs.ScoreDocs)//遍历搜索到的结果 { try { Document doc = searcher.Doc(sd.Doc); string id = doc.Get("id"); string title = doc.Get("title"); string content = doc.Get("content"); string flag = doc.Get("flag"); string imageurl = doc.Get("imageurl"); string updatetime = doc.Get("updatetime"); string createdate = doc.Get("createdate"); PanGu.HighLight.SimpleHTMLFormatter simpleHTMLFormatter = new PanGu.HighLight.SimpleHTMLFormatter("<font color="red">", "</font>"); PanGu.HighLight.Highlighter highlighter = new PanGu.HighLight.Highlighter(simpleHTMLFormatter, new PanGu.Segment()); highlighter.FragmentSize = 50; content = highlighter.GetBestFragment(keyword, content); string titlehighlight = highlighter.GetBestFragment(keyword, title); if (titlehighlight != "") title = titlehighlight; list.Add(new MySearchUnit(id, title, content, flag, imageurl, updatetime)); } catch (Exception ex) { Console.WriteLine(ex.Message); } counter++; } return list; } //st.Stop(); //Response.Write("查询时间:" + st.ElapsedMilliseconds + " 毫秒<br/>"); } #endregion #region 在不同的分类下再根据title和content字段中查询数据(分页) /// <summary> /// 在不同的类型下再根据title和content字段中查询数据(分页) /// </summary> /// <param name="_flag">分类,传空值查询全部</param> /// <param name="keyword"></param> /// <param name="PageIndex"></param> /// <param name="PageSize"></param> /// <param name="TotalCount"></param> /// <returns></returns> public List<MySearchUnit> Search(string _flag, string keyword, int PageIndex, int PageSize, out int TotalCount) { if (PageIndex < 1) PageIndex = 1; //Stopwatch st = new Stopwatch(); //st.Start(); BooleanQuery bq = new BooleanQuery(); if (_flag != "") { QueryParser qpflag = new QueryParser(version, "flag", analyzer); Query qflag = qpflag.Parse(_flag); bq.Add(qflag, Occur.MUST);//与运算 } if (keyword != "") { string[] fileds = { "title", "content" };//查询字段 QueryParser parser = null;// new QueryParser(version, field, analyzer);//一个字段查询 parser = new MultiFieldQueryParser(version, fileds, analyzer);//多个字段查询 Query queryKeyword = parser.Parse(keyword); bq.Add(queryKeyword, Occur.MUST);//与运算 } TopScoreDocCollector collector = TopScoreDocCollector.Create(PageIndex * PageSize, false); IndexSearcher searcher = new IndexSearcher(directory_luce, true);//true-表示只读 searcher.Search(bq, collector); if (collector == null || collector.TotalHits == 0) { TotalCount = 0; return null; } else { int start = PageSize * (PageIndex - 1); //结束数 int limit = PageSize; ScoreDoc[] hits = collector.TopDocs(start, limit).ScoreDocs; List<MySearchUnit> list = new List<MySearchUnit>(); int counter = 1; TotalCount = collector.TotalHits; foreach (ScoreDoc sd in hits)//遍历搜索到的结果 { try { Document doc = searcher.Doc(sd.Doc); string id = doc.Get("id"); string title = doc.Get("title"); string content = doc.Get("content"); string flag = doc.Get("flag"); string imageurl = doc.Get("imageurl"); string updatetime = doc.Get("updatetime"); PanGu.HighLight.SimpleHTMLFormatter simpleHTMLFormatter = new PanGu.HighLight.SimpleHTMLFormatter("<font color="red">", "</font>"); PanGu.HighLight.Highlighter highlighter = new PanGu.HighLight.Highlighter(simpleHTMLFormatter, new PanGu.Segment()); highlighter.FragmentSize = 50; content = highlighter.GetBestFragment(keyword, content); string titlehighlight = highlighter.GetBestFragment(keyword, title); if (titlehighlight != "") title = titlehighlight; list.Add(new MySearchUnit(id, title, content, flag, imageurl, updatetime)); } catch (Exception ex) { Console.WriteLine(ex.Message); } counter++; } return list; } //st.Stop(); //Response.Write("查询时间:" + st.ElapsedMilliseconds + " 毫秒<br/>"); } #endregion #region 删除索引数据(根据id) /// <summary> /// 删除索引数据(根据id) /// </summary> /// <param name="id"></param> /// <returns></returns> public bool Delete(string id) { bool IsSuccess = false; Term term = new Term("id", id); //Analyzer analyzer = new StandardAnalyzer(Lucene.Net.Util.Version.LUCENE_30); //Version version = new Version(); //MultiFieldQueryParser parser = new MultiFieldQueryParser(version, new string[] { "name", "job" }, analyzer);//多个字段查询 //Query query = parser.Parse("小王"); //IndexReader reader = IndexReader.Open(directory_luce, false); //reader.DeleteDocuments(term); //Response.Write("删除记录结果: " + reader.HasDeletions + "<br/>"); //reader.Dispose(); IndexWriter writer = new IndexWriter(directory_luce, analyzer, false, IndexWriter.MaxFieldLength.LIMITED); writer.DeleteDocuments(term); // writer.DeleteDocuments(term)或者writer.DeleteDocuments(query); ////writer.DeleteAll(); writer.Commit(); //writer.Optimize();// IsSuccess = writer.HasDeletions(); writer.Dispose(); return IsSuccess; } #endregion #region 删除全部索引数据 /// <summary> /// 删除全部索引数据 /// </summary> /// <returns></returns> public bool DeleteAll() { bool IsSuccess = true; try { IndexWriter writer = new IndexWriter(directory_luce, analyzer, false, IndexWriter.MaxFieldLength.LIMITED); writer.DeleteAll(); writer.Commit(); //writer.Optimize();// IsSuccess = writer.HasDeletions(); writer.Dispose(); } catch { IsSuccess = false; } return IsSuccess; } #endregion #region directory_luce private Lucene.Net.Store.Directory _directory_luce = null; /// <summary> /// Lucene.Net的目录-参数 /// </summary> public Lucene.Net.Store.Directory directory_luce { get { if (_directory_luce == null) _directory_luce = Lucene.Net.Store.FSDirectory.Open(directory); return _directory_luce; } } #endregion #region directory private System.IO.DirectoryInfo _directory = null; /// <summary> /// 索引在硬盘上的目录 /// </summary> public System.IO.DirectoryInfo directory { get { if (_directory == null) { string dirPath = AppDomain.CurrentDomain.BaseDirectory + "SearchIndex"; if (System.IO.Directory.Exists(dirPath) == false) _directory = System.IO.Directory.CreateDirectory(dirPath); else _directory = new System.IO.DirectoryInfo(dirPath); } return _directory; } } #endregion #region analyzer private Analyzer _analyzer = null; /// <summary> /// 分析器 /// </summary> public Analyzer analyzer { get { //if (_analyzer == null) { _analyzer = new Lucene.Net.Analysis.PanGuAnalyzer();//盘古分词分析器 //_analyzer = new StandardAnalyzer(Lucene.Net.Util.Version.LUCENE_30);//标准分析器 } return _analyzer; } } #endregion #region version private static Lucene.Net.Util.Version _version = Lucene.Net.Util.Version.LUCENE_30; /// <summary> /// 版本号枚举类 /// </summary> public Lucene.Net.Util.Version version { get { return _version; } } #endregion } #region 索引的一个行单元,相当于数据库中的一行数据 /// <summary> /// 索引的一个行单元,相当于数据库中的一行数据 /// </summary> public class MySearchUnit { public MySearchUnit(string _id, string _title, string _content, string _flag, string _imageurl, string _updatetime) { this.id = _id; this.title = _title; this.content = _content; this.flag = _flag; this.imageurl = _imageurl; this.updatetime = _updatetime; } /// <summary> /// 唯一的id号 /// </summary> public string id { get; set; } /// <summary> /// 标题 /// </summary> public string title { get; set; } /// <summary> /// 内容 /// </summary> public string content { get; set; } /// <summary> /// 其他信息 /// </summary> public string flag { get; set; } /// <summary> /// 图片路径 /// </summary> public string imageurl { get; set; } /// <summary> /// 时间 /// </summary> public string updatetime { get; set; } } #endregion }
4、使用 创建索引及搜索
public ActionResult Index() { //PanGuLuceneHelper.instance.DeleteAll();//删除全部 //PanGuLuceneHelper.instance.Delete("1d");//根据id删除 bool exec = false; if (exec) { List<MySearchUnit> list = new List<MySearchUnit>(); list.Add(new MySearchUnit("1a", "标题小王", "今天是小王的生日,大家都很高兴去他家喝酒,玩了一整天。", new Random().Next(1, 10).ToString(), "", "")); list.Add(new MySearchUnit("1b", "标题小张", "今天是小张的生日,大家都很高兴去他家喝酒,玩了几天。", new Random().Next(1, 10).ToString(), "", "")); list.Add(new MySearchUnit("1c", "标题小王", "今天是小王的生日,大家都很高兴去他家喝酒,玩了一整天。", new Random().Next(1, 10).ToString(), "", "")); list.Add(new MySearchUnit("1d", "标题小张", "今天是小张的生日,大家都很高兴去他家喝酒,玩了几天。", new Random().Next(1, 10).ToString(), "", "")); PanGuLuceneHelper.instance.CreateIndex(list);//添加索引 } int count = 0; int PageIndex = 1; int PageSize = 4; string html_content = ""; List<MySearchUnit> searchlist = PanGuLuceneHelper.instance.Search("", "几天", PageIndex, PageSize, out count); html_content += ("查询结果:" + count + "条数据<br/>"); if (searchlist == null || searchlist.Count == 0) { html_content += ("未查询到数据。<br/>"); } else { foreach (MySearchUnit data in searchlist) { html_content += (string.Format("id:{0},title:{1},content:{2},flag:{3},updatetime:{4}<br/>", data.id, data.title, data.content, data.flag, data.updatetime)); } } html_content += (PanGuLuceneHelper.instance.version); return Content(html_content); }