yacy_search_server/source/de/anomic/plasma/plasmaSearchRankingProcess.java

// plasmaSearchRankingProcess.java
// (C) 2007 by Michael Peter Christen; mc@yacy.net, Frankfurt a. M., Germany
// first published 07.11.2007 on http://yacy.net
//
// This is a part of YaCy, a peer-to-peer based web search engine
//
// $LastChangedDate: 2006-04-02 22:40:07 +0200 (So, 02 Apr 2006) $
// $LastChangedRevision: 1986 $
// $LastChangedBy: orbiter $
//
// LICENSE
// 
// This program is free software; you can redistribute it and/or modify
// it under the terms of the GNU General Public License as published by
// the Free Software Foundation; either version 2 of the License, or
// (at your option) any later version.
//
// This program is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
// GNU General Public License for more details.
//
// You should have received a copy of the GNU General Public License
// along with this program; if not, write to the Free Software
// Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA

package de.anomic.plasma;

import java.io.File;
import java.io.IOException;
import java.util.ArrayList;
import java.util.HashMap;
import java.util.Iterator;
import java.util.Map;
import java.util.Set;
import java.util.TreeSet;
import java.util.concurrent.ConcurrentHashMap;

import de.anomic.htmlFilter.htmlFilterContentScraper;
import de.anomic.index.indexContainer;
import de.anomic.index.indexRWIEntry;
import de.anomic.index.indexRWIEntryOrder;
import de.anomic.index.indexRWIVarEntry;
import de.anomic.index.indexURLReference;
import de.anomic.index.indexWord;
import de.anomic.kelondro.kelondroBinSearch;
import de.anomic.kelondro.kelondroMScoreCluster;
import de.anomic.kelondro.kelondroSortStack;
import de.anomic.server.serverCodings;
import de.anomic.server.serverFileUtils;
import de.anomic.server.serverProfiling;
import de.anomic.yacy.yacyURL;

public final class plasmaSearchRankingProcess {
    
    public  static kelondroBinSearch[] ybrTables = null; // block-rank tables
    public  static final int maxYBR = 3; // the lower this value, the faster the search
    private static boolean useYBR = true;
    private static final int maxDoubleDomAll = 20, maxDoubleDomSpecial = 10000;
    
    private final kelondroSortStack<indexRWIVarEntry> stack;
    private final HashMap<String, kelondroSortStack<indexRWIVarEntry>> doubleDomCache; // key = domhash (6 bytes); value = like stack
    private final HashMap<String, String> handover; // key = urlhash, value = urlstring; used for double-check of urls that had been handed over to search process
    private final plasmaSearchQuery query;
    private final int maxentries;
    private int remote_peerCount, remote_indexCount, remote_resourceSize, local_resourceSize;
    private final indexRWIEntryOrder order;
    private final ConcurrentHashMap<String, Integer> urlhashes; // map for double-check; String/Long relation, addresses ranking number (backreference for deletion)
    private final kelondroMScoreCluster<String> ref;  // reference score computation for the commonSense heuristic
    private final int[] flagcount; // flag counter
    private final TreeSet<String> misses; // contains url-hashes that could not been found in the LURL-DB
    private final plasmaWordIndex wordIndex;
    private HashMap<String, indexContainer>[] localSearchContainerMaps;
    private final int[] domZones;
    
    public plasmaSearchRankingProcess(final plasmaWordIndex wordIndex, final plasmaSearchQuery query, final int maxentries, final int concurrency) {
        // we collect the urlhashes and construct a list with urlEntry objects
        // attention: if minEntries is too high, this method will not terminate within the maxTime
        // sortorder: 0 = hash, 1 = url, 2 = ranking
        this.localSearchContainerMaps = null;
        this.stack = new kelondroSortStack<indexRWIVarEntry>(maxentries);
        this.doubleDomCache = new HashMap<String, kelondroSortStack<indexRWIVarEntry>>();
        this.handover = new HashMap<String, String>();
        this.order = (query == null) ? null : new indexRWIEntryOrder(query.ranking);
        this.query = query;
        this.maxentries = maxentries;
        this.remote_peerCount = 0;
        this.remote_indexCount = 0;
        this.remote_resourceSize = 0;
        this.local_resourceSize = 0;
        this.urlhashes = new ConcurrentHashMap<String, Integer>(0, 0.75f, concurrency);
        this.ref = new kelondroMScoreCluster<String>();
        this.misses = new TreeSet<String>();
        this.wordIndex = wordIndex;
        this.flagcount = new int[32];
        for (int i = 0; i < 32; i++) {this.flagcount[i] = 0;}
        this.domZones = new int[8];
        for (int i = 0; i < 8; i++) {this.domZones[i] = 0;}
    }
    
    public long ranking(final indexRWIVarEntry word) {
        return order.cardinal(word);
    }
    
    public int[] zones() {
        return this.domZones;
    }
    
    public void execQuery() {
        
        long timer = System.currentTimeMillis();
        this.localSearchContainerMaps = wordIndex.localSearchContainers(query, null);
        serverProfiling.update("SEARCH", new plasmaProfiling.searchEvent(query.id(true), plasmaSearchEvent.COLLECTION, this.localSearchContainerMaps[0].size(), System.currentTimeMillis() - timer));
        
        // join and exclude the local result
        timer = System.currentTimeMillis();
        final indexContainer index =
            (this.localSearchContainerMaps == null) ?
              plasmaWordIndex.emptyContainer(null, 0) :
                  indexContainer.joinExcludeContainers(
                      this.localSearchContainerMaps[0].values(),
                      this.localSearchContainerMaps[1].values(),
                      query.maxDistance);
        serverProfiling.update("SEARCH", new plasmaProfiling.searchEvent(query.id(true), plasmaSearchEvent.JOIN, index.size(), System.currentTimeMillis() - timer));
        if ((index == null) || (index.size() == 0)) {
            return;
        }
        
        insertRanked(index, true, index.size());
    }
    
    public void insertRanked(final indexContainer index, final boolean local, final int fullResource) {
        // we collect the urlhashes and construct a list with urlEntry objects
        // attention: if minEntries is too high, this method will not terminate within the maxTime

        assert (index != null);
        if (index.size() == 0) return;
        if (local) {
            this.local_resourceSize += fullResource;
        } else {
            this.remote_resourceSize += fullResource;
            this.remote_peerCount++;
        }
        
        long timer = System.currentTimeMillis();
        
        // normalize entries
        final ArrayList<indexRWIVarEntry> decodedEntries = this.order.normalizeWith(index);
        serverProfiling.update("SEARCH", new plasmaProfiling.searchEvent(query.id(true), plasmaSearchEvent.NORMALIZING, index.size(), System.currentTimeMillis() - timer));
        
        // iterate over normalized entries and select some that are better than currently stored
        timer = System.currentTimeMillis();
        final Iterator<indexRWIVarEntry> i = decodedEntries.iterator();
        indexRWIVarEntry iEntry;
        Long r;
        while (i.hasNext()) {
            iEntry = i.next();
            assert (iEntry.urlHash().length() == index.row().primaryKeyLength);
            //if (iEntry.urlHash().length() != index.row().primaryKeyLength) continue;

            // increase flag counts
            for (int j = 0; j < 32; j++) {
                if (iEntry.flags().get(j)) {flagcount[j]++;}
            }
            
            // kick out entries that are too bad according to current findings
            r = Long.valueOf(order.cardinal(iEntry));
            if ((maxentries >= 0) && (stack.size() >= maxentries) && (stack.bottom(r.longValue()))) continue;
            
            // check constraints
            if (!testFlags(iEntry)) continue;
            
            // check document domain
            if (query.contentdom != plasmaSearchQuery.CONTENTDOM_TEXT) {
                if ((query.contentdom == plasmaSearchQuery.CONTENTDOM_AUDIO) && (!(iEntry.flags().get(plasmaCondenser.flag_cat_hasaudio)))) continue;
                if ((query.contentdom == plasmaSearchQuery.CONTENTDOM_VIDEO) && (!(iEntry.flags().get(plasmaCondenser.flag_cat_hasvideo)))) continue;
                if ((query.contentdom == plasmaSearchQuery.CONTENTDOM_IMAGE) && (!(iEntry.flags().get(plasmaCondenser.flag_cat_hasimage)))) continue;
                if ((query.contentdom == plasmaSearchQuery.CONTENTDOM_APP  ) && (!(iEntry.flags().get(plasmaCondenser.flag_cat_hasapp  )))) continue;
            }

            // check tld domain
            if (!yacyURL.matchesAnyDomDomain(iEntry.urlHash(), this.query.zonecode)) {
                // filter out all tld that do not match with wanted tld domain
                continue;
            }
            
            // count domZones
            /*
            indexURLEntry uentry = wordIndex.loadedURL.load(iEntry.urlHash, iEntry, 0); // this eats up a lot of time!!!
            yacyURL uurl = (uentry == null) ? null : uentry.comp().url();
            System.out.println("DEBUG domDomain dom=" + ((uurl == null) ? "null" : uurl.getHost()) + ", zone=" + yacyURL.domDomain(iEntry.urlHash()));
            */
            this.domZones[yacyURL.domDomain(iEntry.urlHash())]++;
            
            // insert
            if ((maxentries < 0) || (stack.size() < maxentries)) {
                // in case that we don't have enough yet, accept any new entry
                if (urlhashes.containsKey(iEntry.urlHash())) continue;
                stack.push(iEntry, r);
            } else {
                // if we already have enough entries, insert only such that are necessary to get a better result
                if (stack.bottom(r.longValue())) {
                    continue;
                }
                // double-check
                if (urlhashes.containsKey(iEntry.urlHash())) continue;
                stack.push(iEntry, r);
            }
            
            // increase counter for statistics
            if (!local) this.remote_indexCount++;
        }
        
        //if ((query.neededResults() > 0) && (container.size() > query.neededResults())) remove(true, true);
        serverProfiling.update("SEARCH", new plasmaProfiling.searchEvent(query.id(true), plasmaSearchEvent.PRESORT, index.size(), System.currentTimeMillis() - timer));
    }

    private boolean testFlags(final indexRWIEntry ientry) {
        if (query.constraint == null) return true;
        // test if ientry matches with filter
        // if all = true: let only entries pass that has all matching bits
        // if all = false: let all entries pass that has at least one matching bit
        if (query.allofconstraint) {
            for (int i = 0; i < 32; i++) {
                if ((query.constraint.get(i)) && (!ientry.flags().get(i))) return false;
            }
            return true;
        }
        for (int i = 0; i < 32; i++) {
            if ((query.constraint.get(i)) && (ientry.flags().get(i))) return true;
        }
        return false;
    }
    
    public Map<String, indexContainer>[] searchContainerMaps() {
        // direct access to the result maps is needed for abstract generation
        // this is only available if execQuery() was called before
        return localSearchContainerMaps;
    }
    
    // todo:
    // - remove redundant urls (sub-path occurred before)
    // - move up shorter urls
    // - root-domain guessing to prefer the root domain over other urls if search word appears in domain name
    
    
    private kelondroSortStack<indexRWIVarEntry>.stackElement bestRWI(final boolean skipDoubleDom) {
        // returns from the current RWI list the best entry and removes this entry from the list
        kelondroSortStack<indexRWIVarEntry> m;
        kelondroSortStack<indexRWIVarEntry>.stackElement rwi;
        while (stack.size() > 0) {
            rwi = stack.pop();
            if (rwi == null) continue; // in case that a synchronization problem occurred just go lazy over it
            if (!skipDoubleDom) return rwi;
            // check doubledom
            final String domhash = rwi.element.urlHash().substring(6);
            m = this.doubleDomCache.get(domhash);
            if (m == null) {
                // first appearance of dom
                m = new kelondroSortStack<indexRWIVarEntry>((query.specialRights) ? maxDoubleDomSpecial : maxDoubleDomAll);
                this.doubleDomCache.put(domhash, m);
                return rwi;
            }
            // second appearances of dom
            m.push(rwi);
        }
        // no more entries in sorted RWI entries. Now take Elements from the doubleDomCache
        // find best entry from all caches
        final Iterator<kelondroSortStack<indexRWIVarEntry>> i = this.doubleDomCache.values().iterator();
        kelondroSortStack<indexRWIVarEntry>.stackElement bestEntry = null;
        kelondroSortStack<indexRWIVarEntry>.stackElement o;
        while (i.hasNext()) {
            m = i.next();
            if (m == null) continue;
            if (m.size() == 0) continue;
            if (bestEntry == null) {
                bestEntry = m.top();
                continue;
            }
            o = m.top();
            if (o.weight.longValue() < bestEntry.weight.longValue()) {
                bestEntry = o;
            }
        }
        if (bestEntry == null) return null;
        // finally remove the best entry from the doubledom cache
        m = this.doubleDomCache.get(bestEntry.element.urlHash().substring(6));
        o = m.pop();
        assert o == null || o.element.urlHash().equals(bestEntry.element.urlHash());
        return bestEntry;
    }
    
    public indexURLReference bestURL(final boolean skipDoubleDom) {
        // returns from the current RWI list the best URL entry and removed this entry from the list
        while ((stack.size() > 0) || (size() > 0)) {
                if (((stack.size() == 0) && (size() == 0))) break;
                final kelondroSortStack<indexRWIVarEntry>.stackElement obrwi = bestRWI(skipDoubleDom);
                if (obrwi == null) continue; // *** ? this happened and the thread was suspended silently. cause?
                final indexURLReference u = wordIndex.getURL(obrwi.element.urlHash(), obrwi.element, obrwi.weight.longValue());
                if (u != null) {
                    final indexURLReference.Components comp = u.comp();
                    if (comp.url() != null) this.handover.put(u.hash(), comp.url().toNormalform(true, false)); // remember that we handed over this url
                    return u;
                }
                misses.add(obrwi.element.urlHash());
        }
        return null;
    }
    
    public int size() {
        //assert sortedRWIEntries.size() == urlhashes.size() : "sortedRWIEntries.size() = " + sortedRWIEntries.size() + ", urlhashes.size() = " + urlhashes.size();
        int c = stack.size();
        final Iterator<kelondroSortStack<indexRWIVarEntry>> i = this.doubleDomCache.values().iterator();
        while (i.hasNext()) c += i.next().size();
        return c;
    }
    
    public int[] flagCount() {
    	return flagcount;
    }
    
    // "results from a total number of <remote_resourceSize + local_resourceSize> known (<local_resourceSize> local, <remote_resourceSize> remote), <remote_indexCount> links from <remote_peerCount> other YaCy peers."
    
    public int filteredCount() {
        // the number of index entries that are considered as result set
        return this.stack.size();
    }

    public int getRemoteIndexCount() {
        // the number of result contributions from all the remote peers
        return this.remote_indexCount;
    }
    
    public int getRemotePeerCount() {
        // the number of remote peers that have contributed
        return this.remote_peerCount;
    }
    
    public int getRemoteResourceSize() {
        // the number of all hits in all the remote peers
        return this.remote_resourceSize;
    }
    
    public int getLocalResourceSize() {
        // the number of hits in the local peer (index size, size of the collection in the own index)
        return this.local_resourceSize;
    }
    
    public indexRWIEntry remove(final String urlHash) {
        final kelondroSortStack<indexRWIVarEntry>.stackElement se = stack.remove(urlHash.hashCode());
        if (se == null) return null;
        urlhashes.remove(urlHash);
        return se.element;
    }
    
    public Iterator<String> miss() {
        return this.misses.iterator();
    }
    
    public Set<String> getReferences(final int count) {
        // create a list of words that had been computed by statistics over all
        // words that appeared in the url or the description of all urls
        final Object[] refs = ref.getScores(count, false, 2, Integer.MAX_VALUE);
        final TreeSet<String> s = new TreeSet<String>(String.CASE_INSENSITIVE_ORDER);
        for (int i = 0; i < refs.length; i++) {
            s.add((String) refs[i]);
        }
        return s;
    }
    
    public void addReferences(final String[] words) {
        String word;
        for (int i = 0; i < words.length; i++) {
            word = words[i].toLowerCase();
            if ((word.length() > 2) &&
                ("http_html_php_ftp_www_com_org_net_gov_edu_index_home_page_for_usage_the_and_".indexOf(word) < 0) &&
                (!(query.queryHashes.contains(indexWord.word2hash(word)))))
                ref.incScore(word);
        }
    }
    
    protected void addReferences(final plasmaSearchEvent.ResultEntry resultEntry) {
        // take out relevant information for reference computation
        if ((resultEntry.url() == null) || (resultEntry.title() == null)) return;
        final String[] urlcomps = htmlFilterContentScraper.urlComps(resultEntry.url().toNormalform(true, true)); // word components of the url
        final String[] descrcomps = resultEntry.title().toLowerCase().split(htmlFilterContentScraper.splitrex); // words in the description
        
        // add references
        addReferences(urlcomps);
        addReferences(descrcomps);
    }
    
    public indexRWIEntryOrder getOrder() {
        return this.order;
    }
    
    public static void loadYBR(final File rankingPath, final int count) {
        // load ranking tables
        if (rankingPath.exists()) {
            ybrTables = new kelondroBinSearch[count];
            String ybrName;
            File f;
            try {
                for (int i = 0; i < count; i++) {
                    ybrName = "YBR-4-" + serverCodings.encodeHex(i, 2) + ".idx";
                    f = new File(rankingPath, ybrName);
                    if (f.exists()) {
                        ybrTables[i] = new kelondroBinSearch(serverFileUtils.read(f), 6);
                    } else {
                        ybrTables[i] = null;
                    }
                }
            } catch (final IOException e) {
                ybrTables = null;
            }
        } else {
            ybrTables = null;
        }
    }
    
    public static boolean canUseYBR() {
        return ybrTables != null;
    }
    
    public static boolean isUsingYBR() {
        return useYBR;
    }
    
    public static void switchYBR(final boolean usage) {
        useYBR = usage;
    }
    
    public static int ybr(final String urlHash) {
        // returns the YBR value in a range of 0..15, where 0 means best ranking and 15 means worst ranking
        if (ybrTables == null) return 15;
        if (!(useYBR)) return 15;
        final String domHash = urlHash.substring(6);
        final int m = Math.min(maxYBR, ybrTables.length);
        for (int i = 0; i < m; i++) {
            if ((ybrTables[i] != null) && (ybrTables[i].contains(domHash.getBytes()))) {
                //System.out.println("YBR FOUND: " + urlHash + " (" + i + ")");
                return i;
            }
        }
        //System.out.println("NOT FOUND: " + urlHash);
        return 15;
    }
    
    public long postRanking(
                    final Set<String> topwords,
                    final plasmaSearchEvent.ResultEntry rentry,
                    final int position) {

        long r = (255 - position) << 8;
        
        // for media search: prefer pages with many links
        if (query.contentdom == plasmaSearchQuery.CONTENTDOM_IMAGE) r += rentry.limage() << query.ranking.coeff_cathasimage;
        if (query.contentdom == plasmaSearchQuery.CONTENTDOM_AUDIO) r += rentry.laudio() << query.ranking.coeff_cathasaudio;
        if (query.contentdom == plasmaSearchQuery.CONTENTDOM_VIDEO) r += rentry.lvideo() << query.ranking.coeff_cathasvideo;
        if (query.contentdom == plasmaSearchQuery.CONTENTDOM_APP  ) r += rentry.lapp()   << query.ranking.coeff_cathasapp;
        
        // prefer hit with 'prefer' pattern
        if (rentry.url().toNormalform(true, true).matches(query.prefer)) r += 256 << query.ranking.coeff_prefer;
        if (rentry.title().matches(query.prefer)) r += 256 << query.ranking.coeff_prefer;
        
        // apply 'common-sense' heuristic using references
        final String urlstring = rentry.url().toNormalform(true, true);
        final String[] urlcomps = htmlFilterContentScraper.urlComps(urlstring);
        final String[] descrcomps = rentry.title().toLowerCase().split(htmlFilterContentScraper.splitrex);
        for (int j = 0; j < urlcomps.length; j++) {
            if (topwords.contains(urlcomps[j])) r += Math.max(1, 256 - urlstring.length()) << query.ranking.coeff_urlcompintoplist;
        }
        for (int j = 0; j < descrcomps.length; j++) {
            if (topwords.contains(descrcomps[j])) r += Math.max(1, 256 - rentry.title().length()) << query.ranking.coeff_descrcompintoplist;
        }

        // apply query-in-result matching
        final Set<String> urlcomph = indexWord.words2hashSet(urlcomps);
        final Set<String> descrcomph = indexWord.words2hashSet(descrcomps);
        final Iterator<String> shi = query.queryHashes.iterator();
        String queryhash;
        while (shi.hasNext()) {
            queryhash = shi.next();
            if (urlcomph.contains(queryhash)) r += 256 << query.ranking.coeff_appurl;
            if (descrcomph.contains(queryhash)) r += 256 << query.ranking.coeff_app_dc_title;
        }

        return r;
    }
}