crawler4j_迷你搜索引擎–使用Neo4j，Crawler4j，Graphstream和Encog的基础知识

crawler4j

继续执行正在实现搜索引擎的Programming Collection Intelligence （PCI）的第4章。

我可能比做一次运动所咬的东西要多。我认为，与其使用本书中所使用的常规关系数据库结构，不如说我一直想看看Neo4J，所以现在是时候了。只是说，这不一定是图数据库的理想用例，但是用1块石头杀死3只鸟可能有多难。

在尝试重置SQL Server的教程中，Oracle的想法花了比预期更长的时间，但是幸运的是Neo4j周围有一些很棒的资源。

只是几个：

neo4j –学习
忙碌的开发人员的图论
Graph数据库

由于我只是想作为一个小练习来运行它，所以我决定采用内存中的实现方式，而不是将其作为服务在我的机器上运行。事后看来，这可能是一个错误，而工具和Web界面将帮助我从一开始就更快地可视化数据图。

因为您只能在内存中实现1个可写实例，所以我做了一个双锁单例工厂来创建和清除数据库。

package net.briandupreez.pci.chapter4;import org.neo4j.graphdb.GraphDatabaseService;
import org.neo4j.graphdb.factory.GraphDatabaseFactory;
import org.neo4j.kernel.impl.util.FileUtils;import java.io.File;
import java.io.IOException;
import java.util.HashMap;
import java.util.Map;public class CreateDBFactory {private static GraphDatabaseService graphDb = null;public static final String RESOURCES_CRAWL_DB = "resources/crawl/db";public static GraphDatabaseService createInMemoryDB() {if (null == graphDb) {synchronized (GraphDatabaseService.class) {if (null == graphDb) {final Map<String, String> config = new HashMap<>();config.put("neostore.nodestore.db.mapped_memory", "50M");config.put("string_block_size", "60");config.put("array_block_size", "300");graphDb = new GraphDatabaseFactory().newEmbeddedDatabaseBuilder(RESOURCES_CRAWL_DB).setConfig(config).newGraphDatabase();registerShutdownHook(graphDb);}}}return graphDb;}private static void registerShutdownHook(final GraphDatabaseService graphDb) {Runtime.getRuntime().addShutdownHook(new Thread() {@Overridepublic void run() {graphDb.shutdown();}});}public static void clearDb() {try {if(graphDb != null){graphDb.shutdown();graphDb = null;}FileUtils.deleteRecursively(new File(RESOURCES_CRAWL_DB));} catch (final IOException e) {throw new RuntimeException(e);}}}

然后使用Crawler4j创建了一个图形，其中包含以我的博客开头的所有URL，它们与其他URL的关系以及这些URL包含的所有单词和单词的索引。

package net.briandupreez.pci.chapter4;import edu.uci.ics.crawler4j.crawler.Page;
import edu.uci.ics.crawler4j.crawler.WebCrawler;
import edu.uci.ics.crawler4j.parser.HtmlParseData;
import edu.uci.ics.crawler4j.url.WebURL;
import org.neo4j.graphdb.GraphDatabaseService;
import org.neo4j.graphdb.Node;
import org.neo4j.graphdb.Relationship;
import org.neo4j.graphdb.Transaction;
import org.neo4j.graphdb.index.Index;import java.util.ArrayList;
import java.util.Arrays;
import java.util.List;public class Neo4JWebCrawler extends WebCrawler {private final GraphDatabaseService graphDb;/*** Constructor.*/public Neo4JWebCrawler() {this.graphDb = CreateDBFactory.createInMemoryDB();}@Overridepublic boolean shouldVisit(final WebURL url) {final String href = url.getURL().toLowerCase();return !NodeConstants.FILTERS.matcher(href).matches();}/*** This function is called when a page is fetched and ready* to be processed by your program.*/@Overridepublic void visit(final Page page) {final String url = page.getWebURL().getURL();System.out.println("URL: " + url);final Index<Node> nodeIndex = graphDb.index().forNodes(NodeConstants.PAGE_INDEX);if (page.getParseData() instanceof HtmlParseData) {HtmlParseData htmlParseData = (HtmlParseData) page.getParseData();String text = htmlParseData.getText();//String html = htmlParseData.getHtml();List<WebURL> links = htmlParseData.getOutgoingUrls();Transaction tx = graphDb.beginTx();try {final Node pageNode = graphDb.createNode();pageNode.setProperty(NodeConstants.URL, url);nodeIndex.add(pageNode, NodeConstants.URL, url);//get all the wordsfinal List<String> words = cleanAndSplitString(text);int index = 0;for (final String word : words) {final Node wordNode = graphDb.createNode();wordNode.setProperty(NodeConstants.WORD, word);wordNode.setProperty(NodeConstants.INDEX, index++);final Relationship relationship = pageNode.createRelationshipTo(wordNode, RelationshipTypes.CONTAINS);relationship.setProperty(NodeConstants.SOURCE, url);}for (final WebURL webURL : links) {System.out.println("Linking to " + webURL);final Node linkNode = graphDb.createNode();linkNode.setProperty(NodeConstants.URL, webURL.getURL());final Relationship relationship = pageNode.createRelationshipTo(linkNode, RelationshipTypes.LINK_TO);relationship.setProperty(NodeConstants.SOURCE, url);relationship.setProperty(NodeConstants.DESTINATION, webURL.getURL());}tx.success();} finally {tx.finish();}}}private static List<String> cleanAndSplitString(final String input) {if (input != null) {final String[] dic = input.toLowerCase().replaceAll("\\p{Punct}", "").replaceAll("\\p{Digit}", "").split("\\s+");return Arrays.asList(dic);}return new ArrayList<>();}}

收集完数据后，我可以查询它并执行搜索引擎的功能。为此，我决定使用Java Futures，因为这是我仅读过但尚未实现的另一件事。在我的日常工作环境中，我们使用应用程序服务器中的Weblogic / CommonJ工作管理器来执行相同的任务。

final ExecutorService executorService = Executors.newFixedThreadPool(4);final String[] searchTerms = {"java", "spring"};List<Callable<TaskResponse>> tasks = new ArrayList<>();tasks.add(new WordFrequencyTask(searchTerms));tasks.add(new DocumentLocationTask(searchTerms));tasks.add(new PageRankTask(searchTerms));tasks.add(new NeuralNetworkTask(searchTerms));final List<Future<TaskResponse>> results = executorService.invokeAll(tasks);

然后，我开始为以下每个任务创建一个任务，对单词频率，文档位置，页面排名和神经网络（带有虚假输入/训练数据）进行计数，以根据搜索条件对返回的页面进行排名。所有代码都在我的公共github博客仓库中。

免责声明：神经网络任务要么没有足够的数据来有效，要么我没有正确实现数据标准化，所以它目前不是很有用，我将在完成while PCI的旅程后再返回书。

值得共享的一项任务是Page Rank，我很快就读懂了一些理论，认为我不那么聪明，然后去寻找实现它的图书馆。我发现Graphstream是一个很棒的开源项目，它不仅可以完成PageRank的全部工作，还可以查看他们的视频。

因此，很容易实现本练习的PageRank任务。

package net.briandupreez.pci.chapter4.tasks;import net.briandupreez.pci.chapter4.NodeConstants;
import net.briandupreez.pci.chapter4.NormalizationFunctions;
import org.graphstream.algorithm.PageRank;
import org.graphstream.graph.Graph;
import org.graphstream.graph.implementations.SingleGraph;
import org.neo4j.cypher.javacompat.ExecutionEngine;
import org.neo4j.cypher.javacompat.ExecutionResult;
import org.neo4j.graphdb.Node;
import org.neo4j.graphdb.Relationship;import java.util.HashMap;
import java.util.Iterator;
import java.util.Map;
import java.util.concurrent.Callable;public class PageRankTask extends SearchTask implements Callable<TaskResponse> {public PageRankTask(final String... terms) {super(terms);}@Overrideprotected ExecutionResult executeQuery(final String... words) {final ExecutionEngine engine = new ExecutionEngine(graphDb);final StringBuilder bob = new StringBuilder("START page=node(*) MATCH (page)-[:CONTAINS]->words ");bob.append(", (page)-[:LINK_TO]->related ");bob.append("WHERE words.word in [");bob.append(formatArray(words));bob.append("] ");bob.append("RETURN DISTINCT page, related");return engine.execute(bob.toString());}public TaskResponse call() {final ExecutionResult result = executeQuery(searchTerms);final Map<String, Double> returnMap = convertToUrlTotalWords(result);final TaskResponse response = new TaskResponse();response.taskClazz = this.getClass();response.resultMap = NormalizationFunctions.normalizeMap(returnMap, true);return response;}private Map<String, Double> convertToUrlTotalWords(final ExecutionResult result) {final Map<String, Double> uniqueUrls = new HashMap<>();final Graph g = new SingleGraph("rank", false, true);final Iterator<Node> pageIterator = result.columnAs("related");while (pageIterator.hasNext()) {final Node node = pageIterator.next();final Iterator<Relationship> relationshipIterator = node.getRelationships().iterator();while (relationshipIterator.hasNext()) {final Relationship relationship = relationshipIterator.next();final String source = relationship.getProperty(NodeConstants.SOURCE).toString();uniqueUrls.put(source, 0.0);final String destination = relationship.getProperty(NodeConstants.DESTINATION).toString();g.addEdge(String.valueOf(node.getId()), source, destination, true);}}computeAndSetPageRankScores(uniqueUrls, g);return uniqueUrls;}/*** Compute score** @param uniqueUrls urls* @param graph      the graph of all links*/private void computeAndSetPageRankScores(final Map<String, Double> uniqueUrls, final Graph graph) {final PageRank pr = new PageRank();pr.init(graph);pr.compute();for (final Map.Entry<String, Double> entry : uniqueUrls.entrySet()) {final double score = 100 * pr.getRank(graph.getNode(entry.getKey()));entry.setValue(score);}}}

在这两者之间，我发现了一种通过Stackoverflow上的值对映射进行排序的出色实现。

package net.briandupreez.pci.chapter4;import java.util.*;public class MapUtil {/*** Sort a map based on values.* The values must be Comparable.** @param map       the map to be sorted* @param ascending in ascending order, or descending if false* @param <K>       key generic* @param <V>       value generic* @return sorted list*/public static <K, V extends Comparable<? super V>> List<Map.Entry<K, V>> entriesSortedByValues(final Map<K, V> map, final boolean ascending) {final List<Map.Entry<K, V>> sortedEntries = new ArrayList<>(map.entrySet());Collections.sort(sortedEntries,new Comparator<Map.Entry<K, V>>() {@Overridepublic int compare(final Map.Entry<K, V> e1, final Map.Entry<K, V> e2) {if (ascending) {return e1.getValue().compareTo(e2.getValue());} else {return e2.getValue().compareTo(e1.getValue());}}});return sortedEntries;}}

用于实现所有这些功能的Maven依赖项

<dependency><groupId>com.google.guava</groupId><artifactId>guava</artifactId><version>14.0.1</version></dependency><dependency><groupId>org.encog</groupId><artifactId>encog-core</artifactId><version>3.2.0-SNAPSHOT</version></dependency><dependency><groupId>edu.uci.ics</groupId><artifactId>crawler4j</artifactId><version>3.5</version><type>jar</type><scope>compile</scope></dependency><dependency><groupId>org.neo4j</groupId><artifactId>neo4j</artifactId><version>1.9</version></dependency><dependency><groupId>org.graphstream</groupId><artifactId>gs-algo</artifactId><version>1.1.2</version></dependency>

现在进入关于PCI…优化的第5章。

参考： 迷你搜索引擎–只是基础知识，它们使用了Zen机构IT博客领域的 JCG合作伙伴 Brian Du Preez的Neo4j，Crawler4j，Graphstream和Encog 。