static void generateSummary( Summary &summary, char *htmlInput, const char *queryStr, const char *urlStr ) {
Xml xml;
ASSERT_TRUE(xml.set(htmlInput, strlen(htmlInput), 0, CT_HTML));
Words words;
ASSERT_TRUE(words.set(&xml, true));
Bits bits;
ASSERT_TRUE(bits.set(&words));
Url url;
url.set(urlStr);
Sections sections;
ASSERT_TRUE(sections.set(&words, &bits, &url, "", CT_HTML));
Query query;
ASSERT_TRUE(query.set2(queryStr, langEnglish, true));
LinkInfo linkInfo;
memset ( &linkInfo , 0 , sizeof(LinkInfo) );
linkInfo.m_lisize = sizeof(LinkInfo);
Title title;
ASSERT_TRUE(title.setTitle(&xml, &words, 80, &query, &linkInfo, &url, NULL, 0, CT_HTML, langEnglish));
Pos pos;
ASSERT_TRUE(pos.set(&words));
Bits bitsForSummary;
ASSERT_TRUE(bitsForSummary.setForSummary(&words));
Phrases phrases;
ASSERT_TRUE(phrases.set(&words, &bits));
Matches matches;
matches.setQuery(&query);
ASSERT_TRUE(matches.set(&words, &phrases, §ions, &bitsForSummary, &pos, &xml, &title, &url, &linkInfo));
summary.setSummary(&xml, &words, §ions, &pos, &query, 180, 3, 3, 180, &url, &matches, title.getTitle(), title.getTitleLen());
}
// returns false if blocked, true otherwise
bool processLoop ( void *state ) {
// get it
State2 *st = (State2 *)state;
// get the tcp socket from the state
TcpSocket *s = st->m_socket;
// get it
XmlDoc *xd = &st->m_xd;
if ( ! xd->m_loaded ) {
// setting just the docid. niceness is 0.
//xd->set3 ( st->m_docId , st->m_coll , 0 );
// callback
xd->setCallback ( state , processLoop );
// . and tell it to load from the old title rec
// . this sets xd->m_oldTitleRec/m_oldTitleRecSize
// . this sets xd->ptr_* and all other member vars from
// the old title rec if found in titledb.
if ( ! xd->loadFromOldTitleRec ( ) ) return false;
}
if ( g_errno ) return sendErrorReply ( st , g_errno );
// now force it to load old title rec
//char **tr = xd->getTitleRec();
SafeBuf *tr = xd->getTitleRecBuf();
// blocked? return false if so. it will call processLoop() when it rets
if ( tr == (void *)-1 ) return false;
// we did not block. check for error? this will free "st" too.
if ( ! tr ) return sendErrorReply ( st , g_errno );
// if title rec was empty, that is a problem
if ( xd->m_titleRecBuf.length() == 0 )
return sendErrorReply ( st , ENOTFOUND);
// set callback
char *na = xd->getIsNoArchive();
// wait if blocked
if ( na == (void *)-1 ) return false;
// error?
if ( ! na ) return sendErrorReply ( st , g_errno );
// forbidden? allow turkeys through though...
if ( ! st->m_isAdmin && *na )
return sendErrorReply ( st , ENOCACHE );
SafeBuf *sb = &st->m_sb;
// &page=4 will print rainbow sections
if ( ! st->m_printed && st->m_r.getLong("page",0) ) {
// do not repeat this call
st->m_printed = true;
// this will call us again since we called
// xd->setCallback() above to us
if ( ! xd->printDocForProCog ( sb , &st->m_r ) )
return false;
}
char *contentType = "text/html";
char format = st->m_format;
if ( format == FORMAT_XML ) contentType = "text/xml";
if ( format == FORMAT_JSON ) contentType = "application/json";
// if we printed a special page (like rainbow sections) then return now
if ( st->m_printed ) {
bool status = g_httpServer.sendDynamicPage (s,
//buf,bufLen,
sb->getBufStart(),
sb->getLength(),
-1,false,
//"text/html",
contentType,
-1, NULL, "utf8" );
// nuke state2
mdelete ( st , sizeof(State2) , "PageGet1" );
delete (st);
return status;
}
/*
// this was calling XmlDoc and setting sections, etc. to
// get the SpiderReply junk... no no no
// is it banned or filtered? this ignores the TagRec in the titleRec
// and uses msg8a to get it fresh instead
char *vi = xd->getIsFiltered();//Visible( );
// wait if blocked
if ( vi == (void *)-1 ) return false;
// error?
if ( ! vi ) return sendErrorReply ( st , g_errno );
// banned?
if ( ! st->m_isAdmin && ! *vi ) return sendErrorReply (st,EDOCBANNED);
*/
// get the utf8 content
char **utf8 = xd->getUtf8Content();
//long len = xd->size_utf8Content - 1;
// wait if blocked???
if ( utf8 == (void *)-1 ) return false;
// strange
if ( xd->size_utf8Content<=0) {
log("pageget: utf8 content <= 0");
return sendErrorReply(st,EBADENGINEER );
}
// alloc error?
if ( ! utf8 ) return sendErrorReply ( st , g_errno );
// get this host
Host *h = g_hostdb.getHost ( g_hostdb.m_hostId );
if ( ! h ) {
log("pageget: hostid %li is bad",g_hostdb.m_hostId);
return sendErrorReply(st,EBADENGINEER );
}
char *content = xd->ptr_utf8Content;
long contentLen = xd->size_utf8Content - 1;
// shortcut
char strip = st->m_strip;
// alloc buffer now
//char *buf = NULL;
//long bufMaxSize = 0;
//bufMaxSize = len + ( 32 * 1024 ) ;
//bufMaxSize = contentLen + ( 32 * 1024 ) ;
//buf = (char *)mmalloc ( bufMaxSize , "PageGet2" );
//char *p = buf;
//char *bufEnd = buf + bufMaxSize;
//if ( ! buf ) {
// return sendErrorReply ( st , g_errno );
//}
// for undoing the header
//char *start1 = p;
long startLen1 = sb->length();
// we are always utfu
if ( strip != 2 )
sb->safePrintf( "<meta http-equiv=\"Content-Type\" "
"content=\"text/html;charset=utf8\">\n");
// base href
//Url *base = &xd->m_firstUrl;
//if ( xd->ptr_redirUrl.m_url[0] )
// base = &xd->m_redirUrl;
char *base = xd->ptr_firstUrl;
if ( xd->ptr_redirUrl ) base = xd->ptr_redirUrl;
//Url *redir = *xd->getRedirUrl();
if ( strip != 2 ) {
sb->safePrintf ( "<BASE HREF=\"%s\">" , base );
//p += gbstrlen ( p );
}
// default colors in case css files missing
if ( strip != 2 ) {
sb->safePrintf( "\n<style type=\"text/css\">\n"
"body{background-color:white;color:black;}\n"
"</style>\n");
//p += gbstrlen ( p );
}
//char format = st->m_format;
if ( format == FORMAT_XML ) sb->reset();
if ( format == FORMAT_JSON ) sb->reset();
// for undoing the stuff below
long startLen2 = sb->length();//p;
// query should be NULL terminated
char *q = st->m_q;
long qlen = st->m_qlen;
char styleTitle[128] = "font-size:14px;font-weight:600;"
"color:#000000;";
char styleText[128] = "font-size:14px;font-weight:400;"
"color:#000000;";
char styleLink[128] = "font-size:14px;font-weight:400;"
"color:#0000ff;";
char styleTell[128] = "font-size:14px;font-weight:600;"
"color:#cc0000;";
// get the url of the title rec
Url *f = xd->getFirstUrl();
bool printDisclaimer = st->m_printDisclaimer;
if ( xd->m_contentType == CT_JSON )
printDisclaimer = false;
if ( format == FORMAT_XML ) printDisclaimer = false;
if ( format == FORMAT_JSON ) printDisclaimer = false;
char tbuf[100];
tbuf[0] = 0;
time_t lastSpiderDate = xd->m_spideredTime;
if ( printDisclaimer ||
format == FORMAT_XML ||
format == FORMAT_JSON ) {
struct tm *timeStruct = gmtime ( &lastSpiderDate );
strftime ( tbuf, 100,"%b %d, %Y UTC", timeStruct);
}
// We should always be displaying this disclaimer.
// - May eventually want to display this at a different location
// on the page, or on the click 'n' scroll browser page itself
// when this page is not being viewed solo.
// CNS: if ( ! st->m_clickNScroll ) {
if ( printDisclaimer ) {
sb->safePrintf(//sprintf ( p ,
//"<BASE HREF=\"%s\">"
//"<table border=1 width=100%%>"
//"<tr><td>"
"<table border=\"1\" bgcolor=\"#"
BGCOLOR
"\" cellpadding=\"10\" "
//"id=\"gbcnsdisctable\" class=\"gbcnsdisctable_v\""
"cellspacing=\"0\" width=\"100%%\" color=\"#ffffff\">"
"<tr"
//" id=\"gbcnsdisctr\" class=\"gbcnsdisctr_v\""
"><td>"
//"<font face=times,sans-serif color=black size=-1>"
"<span style=\"%s\">"
"This is Gigablast's cached page of </span>"
"<a href=\"%s\" style=\"%s\">%s</a>"
"" , styleTitle, f->getUrl(), styleLink,
f->getUrl() );
//p += gbstrlen ( p );
// then the rest
//sprintf(p ,
sb->safePrintf(
"<span style=\"%s\">. "
"Gigablast is not responsible for the content of "
"this page.</span>", styleTitle );
//p += gbstrlen ( p );
sb->safePrintf ( "<br/><span style=\"%s\">"
"Cached: </span>"
"<span style=\"%s\">",
styleTitle, styleText );
//p += gbstrlen ( p );
// then the spider date in GMT
// time_t lastSpiderDate = xd->m_spideredTime;
// struct tm *timeStruct = gmtime ( &lastSpiderDate );
// char tbuf[100];
// strftime ( tbuf, 100,"%b %d, %Y UTC", timeStruct);
//p += gbstrlen ( p );
sb->safeStrcpy(tbuf);
// Moved over from PageResults.cpp
sb->safePrintf( "</span> - <a href=\""
"/get?"
"q=%s&c=%s&rtq=%li&"
"d=%lli&strip=1\""
" style=\"%s\">"
"[stripped]</a>",
q , st->m_coll ,
(long)st->m_rtq,
st->m_docId, styleLink );
// a link to alexa
if ( f->getUrlLen() > 5 ) {
sb->safePrintf( " - <a href=\"http:"
"//web.archive.org/web/*/%s\""
" style=\"%s\">"
"[older copies]</a>" ,
f->getUrl(), styleLink );
}
if (st->m_noArchive){
sb->safePrintf( " - <span style=\"%s\"><b>"
"[NOARCHIVE]</b></span>",
styleTell );
}
if (st->m_isBanned){
sb->safePrintf(" - <span style=\"%s\"><b>"
"[BANNED]</b></span>",
styleTell );
}
// only print this if we got a query
if ( qlen > 0 ) {
sb->safePrintf("<br/><br/><span style=\"%s\"> "
"These search terms have been "
"highlighted: ",
styleText );
//p += gbstrlen ( p );
}
}
// how much space left in p?
//long avail = bufEnd - p;
// . make the url that we're outputting for (like in PageResults.cpp)
// . "thisUrl" is the baseUrl for click & scroll
char thisUrl[MAX_URL_LEN];
char *thisUrlEnd = thisUrl + MAX_URL_LEN;
char *x = thisUrl;
// . use the external ip of our gateway
// . construct the NAT mapped port
// . you should have used iptables to map port to the correct
// internal ip:port
//unsigned long ip =g_conf.m_mainExternalIp ; // h->m_externalIp;
//unsigned short port=g_conf.m_mainExternalPort;//h->m_externalHttpPort
// local check
//if ( st->m_isLocal ) {
unsigned long ip = h->m_ip;
unsigned short port = h->m_httpPort;
//}
//sprintf ( x , "http://%s:%li/get?q=" , iptoa ( ip ) , port );
// . we no longer put the port in here
// . but still need http:// since we use <base href=>
if (port == 80) sprintf(x,"http://%s/get?q=",iptoa(ip));
else sprintf(x,"http://%s:%hu/get?q=",iptoa(ip),port);
x += gbstrlen ( x );
// the query url encoded
long elen = urlEncode ( x , thisUrlEnd - x , q , qlen );
x += elen;
// separate cgi vars with a &
//sprintf ( x, "&seq=%li&rtq=%lid=%lli",
// (long)st->m_seq,(long)st->m_rtq,st->m_msg22.getDocId());
sprintf ( x, "&d=%lli",st->m_docId );
x += gbstrlen(x);
// set our query for highlighting
Query qq;
qq.set2 ( q, st->m_langId , true );
// print the query terms into our highlight buffer
Highlight hi;
// make words so we can set the scores to ignore fielded terms
Words qw;
qw.set ( q , // content being highlighted, utf8
qlen , // content being highlighted, utf8
TITLEREC_CURRENT_VERSION,
true , // computeIds
false ); // hasHtmlEntities?
// . assign scores of 0 to query words that should be ignored
// . TRICKY: loop over words in qq.m_qwords, but they should be 1-1
// with words in qw.
// . sanity check
//if ( qw.getNumWords() != qq.m_numWords ) { char *xx = NULL; *xx = 0;}
// declare up here
Matches m;
// do the loop
//Scores ss;
//ss.set ( &qw , NULL );
//for ( long i = 0 ; i < qq.m_numWords ; i++ )
// if ( ! m.matchWord ( &qq.m_qwords[i],i ) ) ss.m_scores[i] = 0;
// now set m.m_matches[] to those words in qw that match a query word
// or phrase in qq.
m.setQuery ( &qq );
//m.addMatches ( &qw , &ss , true );
m.addMatches ( &qw );
long hilen = 0;
// CNS: if ( ! st->m_clickNScroll ) {
// and highlight the matches
if ( printDisclaimer ) {
hilen = hi.set ( //p ,
//avail ,
sb ,
&qw , // words to highlight
&m , // matches relative to qw
false , // doSteming
false , // st->m_clickAndScroll ,
(char *)thisUrl );// base url for ClcknScrll
//p += hilen;
// now an hr
//memcpy ( p , "</span></table></table>\n" , 24 ); p += 24;
sb->safeStrcpy("</span></table></table>\n");
}
bool includeHeader = st->m_includeHeader;
// do not show header for json object display
if ( xd->m_contentType == CT_JSON )
includeHeader = false;
if ( format == FORMAT_XML ) includeHeader = false;
if ( format == FORMAT_JSON ) includeHeader = false;
//mfree(uq, uqCapacity, "PageGet");
// undo the header writes if we should
if ( ! includeHeader ) {
// including base href is off by default when not including
// the header, so the caller must explicitly turn it back on
if ( st->m_includeBaseHref ) sb->m_length=startLen2;//p=start2;
else sb->m_length=startLen1;//p=start1;
}
//sb->safeStrcpy(tbuf);
if ( format == FORMAT_XML ) {
sb->safePrintf("<response>\n");
sb->safePrintf("<statusCode>0</statusCode>\n");
sb->safePrintf("<statusMsg>Success</statusMsg>\n");
sb->safePrintf("<url><![CDATA[");
sb->cdataEncode(xd->m_firstUrl.m_url);
sb->safePrintf("]]></url>\n");
sb->safePrintf("<docId>%llu</docId>\n",xd->m_docId);
sb->safePrintf("\t<cachedTimeUTC>%lu</cachedTimeUTC>\n",
lastSpiderDate);
sb->safePrintf("\t<cachedTimeStr>%s</cachedTimeStr>\n",tbuf);
}
if ( format == FORMAT_JSON ) {
sb->safePrintf("{\"response\":{\n");
sb->safePrintf("\t\"statusCode\":0,\n");
sb->safePrintf("\t\"statusMsg\":\"Success\",\n");
sb->safePrintf("\t\"url\":\"");
sb->jsonEncode(xd->m_firstUrl.m_url);
sb->safePrintf("\",\n");
sb->safePrintf("\t\"docId\":%llu,\n",xd->m_docId);
sb->safePrintf("\t\"cachedTimeUTC\":%lu,\n",lastSpiderDate);
sb->safePrintf("\t\"cachedTimeStr\":\"%s\",\n",tbuf);
}
// identify start of <title> tag we wrote out
char *sbstart = sb->getBufStart();
char *sbend = sb->getBufEnd();
char *titleStart = NULL;
char *titleEnd = NULL;
for ( char *t = sbstart ; t < sbend ; t++ ) {
// title tag?
if ( t[0]!='<' ) continue;
if ( to_lower_a(t[1])!='t' ) continue;
if ( to_lower_a(t[2])!='i' ) continue;
if ( to_lower_a(t[3])!='t' ) continue;
if ( to_lower_a(t[4])!='l' ) continue;
if ( to_lower_a(t[5])!='e' ) continue;
// point to it
char *x = t + 5;
// max - to keep things fast
char *max = x + 500;
for ( ; *x && *x != '>' && x < max ; x++ );
x++;
// find end
char *e = x;
for ( ; *e && e < max ; e++ ) {
if ( e[0]=='<' &&
to_lower_a(e[1])=='/' &&
to_lower_a(e[2])=='t' &&
to_lower_a(e[3])=='i' &&
to_lower_a(e[4])=='t' &&
to_lower_a(e[5])=='l' &&
to_lower_a(e[6])=='e' )
break;
}
if ( e < max ) {
titleStart = x;
titleEnd = e;
}
break;
}
// . print title at top!
// . consider moving
if ( titleStart ) {
char *ebuf = st->m_r.getString("eb");
if ( ! ebuf ) ebuf = "";
//p += sprintf ( p ,
sb->safePrintf(
"<table border=1 "
"cellpadding=10 "
"cellspacing=0 "
"width=100%% "
"color=#ffffff>" );
long printLinks = st->m_r.getLong("links",0);
if ( ! printDisclaimer && printLinks )
sb->safePrintf(//p += sprintf ( p ,
// first put cached and live link
"<tr>"
"<td bgcolor=lightyellow>"
// print cached link
//"<center>"
" "
"<b>"
"<a "
"style=\"font-size:18px;font-weight:600;"
"color:#000000;\" "
"href=\""
"/get?"
"c=%s&d=%lli&qh=0&cnsp=1&eb=%s\">"
"cached link</a>"
" "
"<a "
"style=\"font-size:18px;font-weight:600;"
"color:#000000;\" "
"href=%s>live link</a>"
"</b>"
//"</center>"
"</td>"
"</tr>\n"
,st->m_coll
,st->m_docId
,ebuf
,thisUrl // st->ptr_ubuf
);
if ( printLinks ) {
sb->safePrintf(//p += sprintf ( p ,
"<tr><td bgcolor=pink>"
"<span style=\"font-size:18px;"
"font-weight:600;"
"color:#000000;\">"
" "
"<b>PAGE TITLE:</b> "
);
long tlen = titleEnd - titleStart;
sb->safeMemcpy ( titleStart , tlen );
sb->safePrintf ( "</span></td></tr>" );
}
sb->safePrintf( "</table><br>\n" );
}
// is the content preformatted?
bool pre = false;
char ctype = (char)xd->m_contentType;
if ( ctype == CT_TEXT ) pre = true ; // text/plain
if ( ctype == CT_DOC ) pre = true ; // filtered msword
if ( ctype == CT_PS ) pre = true ; // filtered postscript
if ( format == FORMAT_XML ) pre = false;
if ( format == FORMAT_JSON ) pre = false;
// if it is content-type text, add a <pre>
if ( pre ) {//p + 5 < bufEnd && pre ) {
sb->safePrintf("<pre>");
//p += 5;
}
if ( st->m_strip == 1 )
contentLen = stripHtml( content, contentLen,
(long)xd->m_version, st->m_strip );
// it returns -1 and sets g_errno on error, line OOM
if ( contentLen == -1 ) {
//if ( buf ) mfree ( buf , bufMaxSize , "PageGet2" );
return sendErrorReply ( st , g_errno );
}
Xml xml;
Words ww;
// if no highlighting, skip it
bool queryHighlighting = st->m_queryHighlighting;
if ( st->m_strip == 2 ) queryHighlighting = false;
// do not do term highlighting if json
if ( xd->m_contentType == CT_JSON )
queryHighlighting = false;
SafeBuf tmp;
SafeBuf *xb = sb;
if ( format == FORMAT_XML ) xb = &tmp;
if ( format == FORMAT_JSON ) xb = &tmp;
if ( ! queryHighlighting ) {
xb->safeMemcpy ( content , contentLen );
//p += contentLen ;
}
else {
// get the content as xhtml (should be NULL terminated)
//Words *ww = xd->getWords();
if ( ! xml.set ( content , contentLen , false ,
0 , false , TITLEREC_CURRENT_VERSION ,
false , 0 , CT_HTML ) ) { // niceness is 0
//if ( buf ) mfree ( buf , bufMaxSize , "PageGet2" );
return sendErrorReply ( st , g_errno );
}
if ( ! ww.set ( &xml , true , 0 ) ) { // niceness is 0
//if ( buf ) mfree ( buf , bufMaxSize , "PageGet2" );
return sendErrorReply ( st , g_errno );
}
// sanity check
//if ( ! xd->m_wordsValid ) { char *xx=NULL;*xx=0; }
// how much space left in p?
//avail = bufEnd - p;
Matches m;
m.setQuery ( &qq );
m.addMatches ( &ww );
hilen = hi.set ( xb , // p , avail ,
&ww , &m ,
false /*doStemming?*/ ,
st->m_clickAndScroll ,
thisUrl /*base url for click & scroll*/);
//p += hilen;
log(LOG_DEBUG, "query: Done highlighting cached page content");
}
if ( format == FORMAT_XML ) {
sb->safePrintf("\t<content><![CDATA[");
sb->cdataEncode ( xb->getBufStart() );
sb->safePrintf("]]></content>\n");
sb->safePrintf("</response>\n");
}
if ( format == FORMAT_JSON ) {
sb->safePrintf("\t\"content\":\"\n");
sb->jsonEncode ( xb->getBufStart() );
sb->safePrintf("\"\n}\n}\n");
}
// if it is content-type text, add a </pre>
if ( pre ) { // p + 6 < bufEnd && pre ) {
sb->safeMemcpy ( "</pre>" , 6 );
//p += 6;
}
// calculate bufLen
//long bufLen = p - buf;
long ct = xd->m_contentType;
// now filter the entire buffer to escape out the xml tags
// so it is displayed nice
SafeBuf newbuf;
if ( ct == CT_XML ) {
// encode the xml tags into <tagname> sequences
if ( !newbuf.htmlEncodeXmlTags ( sb->getBufStart() ,
sb->getLength(),
0)){// niceness=0
//if ( buf ) mfree ( buf , bufMaxSize , "PageGet2" );
return sendErrorReply ( st , g_errno );
}
// free out buffer that we alloc'd before returning since this
// should have copied it into another buffer
//if ( buf ) mfree ( buf , bufMaxSize , "PageGet2" );
// reassign
//buf = newbuf.getBufStart();
//bufLen = newbuf.length();
sb->stealBuf ( &newbuf );
}
// now encapsulate it in html head/tail and send it off
// sendErr:
contentType = "text/html";
if ( strip == 2 ) contentType = "text/xml";
// xml is usually buggy and this throws browser off
//if ( ctype == CT_XML ) contentType = "text/xml";
if ( xd->m_contentType == CT_JSON )
contentType = "application/json";
if ( format == FORMAT_XML ) contentType = "text/xml";
if ( format == FORMAT_JSON ) contentType = "application/json";
// safebuf, sb, is a member of "st" so this should copy the buffer
// when it constructs the http reply, and we gotta call delete(st)
// AFTER this so sb is still valid.
bool status = g_httpServer.sendDynamicPage (s,
//buf,bufLen,
sb->getBufStart(),
sb->getLength(),
-1,false,
contentType,
-1, NULL, "utf8" );
// nuke state2
mdelete ( st , sizeof(State2) , "PageGet1" );
delete (st);
// free out buffer that we alloc'd before returning since this
// should have copied it into another buffer
//if ( ct == CT_XML ) newbuf.purge();
//else if ( buf ) mfree ( buf , bufMaxSize , "PageGet2" );
// and convey the status
return status;
}
// returns false and sets g_errno on error
bool Title::setTitle ( Xml *xml, Words *words, int32_t maxTitleLen, Query *query,
LinkInfo *linkInfo, Url *firstUrl, const char *filteredRootTitleBuf, int32_t filteredRootTitleBufSize,
uint8_t contentType, uint8_t langId, int32_t niceness ) {
// make Msg20.cpp faster if it is just has
// Msg20Request::m_setForLinkInfo set to true, no need to extricate a title.
if ( maxTitleLen <= 0 ) {
return true;
}
m_niceness = niceness;
m_maxTitleLen = maxTitleLen;
// if this is too big the "first line" algo can be huge!!!
// and really slow everything way down with a huge title candidate
int32_t maxTitleWords = 128;
// assume no title
reset();
int32_t NW = words->getNumWords();
//
// now get all the candidates
//
// . allow up to 100 title CANDIDATES
// . "as" is the word # of the first word in the candidate
// . "bs" is the word # of the last word IN the candidate PLUS ONE
int32_t n = 0;
int32_t as[MAX_TIT_CANDIDATES];
int32_t bs[MAX_TIT_CANDIDATES];
float scores[MAX_TIT_CANDIDATES];
Words *cptrs[MAX_TIT_CANDIDATES];
int32_t types[MAX_TIT_CANDIDATES];
int32_t parent[MAX_TIT_CANDIDATES];
// record the scoring algos effects
float baseScore [MAX_TIT_CANDIDATES];
float noCapsBoost [MAX_TIT_CANDIDATES];
float qtermsBoost [MAX_TIT_CANDIDATES];
float inCommonCandBoost[MAX_TIT_CANDIDATES];
// reset these
for ( int32_t i = 0 ; i < MAX_TIT_CANDIDATES ; i++ ) {
// assume no parent
parent[i] = -1;
}
// xml and words class for each link info, rss item
Xml tx[MAX_TIT_CANDIDATES];
Words tw[MAX_TIT_CANDIDATES];
int32_t ti = 0;
// restrict how many link texts and rss blobs we check for titles
// because title recs like www.google.com have hundreds and can
// really slow things down to like 50ms for title generation
int32_t kcount = 0;
int32_t rcount = 0;
//int64_t x = gettimeofdayInMilliseconds();
// . get every link text
// . TODO: repeat for linkInfo2, the imported link text
for ( Inlink *k = NULL; linkInfo && (k = linkInfo->getNextInlink(k)) ; ) {
// breathe
QUICKPOLL(m_niceness);
// fast skip check for link text
if ( k->size_linkText >= 3 && ++kcount >= 20 ) continue;
// fast skip check for rss item
if ( k->size_rssItem > 10 && ++rcount >= 20 ) continue;
// set Url
Url u;
u.set( k->getUrl(), k->size_urlBuf );
// is it the same host as us?
bool sh = true;
// skip if not from same host and should be
if ( firstUrl->getHostLen() != u.getHostLen() ) {
sh = false;
}
// skip if not from same host and should be
if ( strncmp( firstUrl->getHost(), u.getHost(), u.getHostLen() ) ) {
sh = false;
}
// get the link text
if ( k->size_linkText >= 3 ) {
char *p = k->getLinkText();
int32_t plen = k->size_linkText - 1;
if ( ! verifyUtf8 ( p , plen ) ) {
log("title: set4 bad link text from url=%s", k->getUrl());
continue;
}
// now the words.
if ( !tw[ti].set( k->getLinkText(), k->size_linkText - 1, true, 0 ) ) {
return false;
}
// set the bookends, it is the whole thing
cptrs [n] = &tw[ti];
as [n] = 0;
bs [n] = tw[ti].getNumWords();
// score higher if same host
if ( sh ) scores[n] = 1.05;
// do not count so high if remote!
else scores[n] = 0.80;
// set the type
if ( sh ) types [n] = TT_LINKTEXTLOCAL;
else types [n] = TT_LINKTEXTREMOTE;
// another candidate
n++;
// use xml and words
ti++;
// break out if too many already. save some for below.
if ( n + 30 >= MAX_TIT_CANDIDATES ) break;
}
// get the rss item
if ( k->size_rssItem <= 10 ) continue;
// . returns false and sets g_errno on error
// . use a 0 for niceness
if ( ! k->setXmlFromRSS ( &tx[ti] , 0 ) ) return false;
// get the word range
int32_t tslen;
bool isHtmlEnc;
char *ts = tx[ti].getRSSTitle ( &tslen , &isHtmlEnc );
// skip if not in the rss
if ( ! ts ) continue;
// skip if empty
if ( tslen <= 0 ) continue;
// now set words to that
if ( !tw[ti].set( ts, tslen, true, 0 ) ) {
return false;
}
// point to that
cptrs [n] = &tw[ti];
as [n] = 0;
bs [n] = tw[ti].getNumWords();
// increment since we are using it
ti++;
// base score for rss title
if ( sh ) scores[n] = 5.0;
// if not same host, treat like link text
else scores[n] = 2.0;
// set the type
if ( sh ) types [n] = TT_RSSITEMLOCAL;
else types [n] = TT_RSSITEMREMOTE;
// advance
n++;
// break out if too many already. save some for below.
if ( n + 30 >= MAX_TIT_CANDIDATES ) break;
}
//logf(LOG_DEBUG,"title: took1=%" PRId64,gettimeofdayInMilliseconds()-x);
//x = gettimeofdayInMilliseconds();
// . set the flags array
// . indicates what words are in title candidates already, but
// that is set below
// . up here we set words that are not allowed to be in candidates,
// like words that are in a link that is not a self link
// . alloc for it
char *flags = NULL;
char localBuf[10000];
int32_t need = words->getNumWords();
if ( need <= 10000 ) {
flags = (char *)localBuf;
} else {
flags = (char *)mmalloc(need,"TITLEflags");
}
if ( ! flags ) {
return false;
}
// clear it
memset ( flags , 0 , need );
// check tags in body
nodeid_t *tids = words->getTagIds();
// scan to set link text flags
// loop over all "words" in the html body
char inLink = false;
char selfLink = false;
for ( int32_t i = 0 ; i < NW ; i++ ) {
// breathe
QUICKPOLL(m_niceness);
// if in a link that is not self link, cannot be in a candidate
if ( inLink && ! selfLink ) {
flags[i] |= 0x02;
}
// out of a link
if ( tids[i] == (TAG_A | BACKBIT) ) {
inLink = false;
}
// if not start of <a> tag, skip it
if ( tids[i] != TAG_A ) {
continue;
}
// flag it
inLink = true;
// get the node in the xml
int32_t xn = words->getNodes()[i];
// is it a self link?
int32_t len;
char *link = xml->getString(xn,"href",&len);
// . set the url class to this
// . TODO: use the base url in the doc
Url u;
u.set( link, len, true, false );
// compare
selfLink = u.equals ( firstUrl );
// skip if not selfLink
if ( ! selfLink ) {
continue;
}
// if it is a selflink , check for an "onClick" tag in the
// anchor tag to fix that Mixx issue for:
// http://www.npr.org/templates/story/story.php?storyId=5417137
int32_t oclen;
char *oc = xml->getString(xn,"onclick",&oclen);
if ( ! oc ) {
oc = xml->getString(xn,"onClick",&oclen);
}
// assume not a self link if we see that...
if ( oc ) {
selfLink = false;
}
// if this <a href> link has a "title" attribute, use that
// instead! that thing is solid gold.
int32_t atlen;
char *atitle = xml->getString(xn,"title",&atlen);
// stop and use that, this thing is gold!
if ( ! atitle || atlen <= 0 ) {
continue;
}
// craziness? ignore it...
if ( atlen > 400 ) {
continue;
}
// if it contains permanent, permalink or share, ignore it!
if ( strncasestr ( atitle, "permalink", atlen ) ||
strncasestr ( atitle,"permanent", atlen) ||
strncasestr ( atitle,"share", atlen) ) {
continue;
}
// do not count the link text as viable
selfLink = false;
// aw, dammit
if ( ti >= MAX_TIT_CANDIDATES ) {
continue;
}
// other dammit
if ( n >= MAX_TIT_CANDIDATES ) {
break;
}
// ok, process it
if ( ! tw[ti].set ( atitle, atlen, true, 0 )) {
return false;
}
// set the bookends, it is the whole thing
cptrs [n] = &tw[ti];
as [n] = 0;
bs [n] = tw[ti].getNumWords();
scores [n] = 3.0; // not ALWAYS solid gold!
types [n] = TT_TITLEATT;
// we are using the words class
ti++;
// advance
n++;
// break out if too many already. save some for below.
if ( n + 20 >= MAX_TIT_CANDIDATES ) {
break;
}
}
//logf(LOG_DEBUG,"title: took2=%" PRId64,gettimeofdayInMilliseconds()-x);
//x = gettimeofdayInMilliseconds();
//int64_t *wids = WW->getWordIds();
// . find the last positive scoring guy
// . do not consider title candidates after "r" if "r" is non-zero
// . FIXES http://larvatusprodeo.net/2009/01/07/partisanship-politics-and-participation/
// the candidate # of the title tag
int32_t tti = -1;
// allow up to 4 tags from each type
char table[512];
// sanity check
if ( getNumXmlNodes() > 512 ) { char *xx=NULL;*xx=0; }
// clear table counts
memset ( table , 0 , 512 );
// the first word
char *wstart = NULL;
if ( NW > 0 ) {
wstart = words->getWord(0);
}
// loop over all "words" in the html body
for ( int32_t i = 0 ; i < NW ; i++ ) {
// come back up here if we encounter another "title-ish" tag
// within our first alleged "title-ish" tag
subloop:
// stop after 30k of text
if ( words->getWord(i) - wstart > 200000 ) {
break; // 1106
}
// get the tag id minus the back tag bit
nodeid_t tid = tids[i] & BACKBITCOMP;
// pen up and pen down for these comment like tags
if ( tid == TAG_SCRIPT || tid == TAG_STYLE ) {
// ignore "titles" in script or style tags
if ( ! (tids[i] & BACKBIT) ) {
continue;
}
}
/// @todo ALC we should allow more tags than just title/link
// skip if not a good tag.
if (tid != TAG_TITLE && tid != TAG_A) {
continue;
}
// must NOT be a back tag
if ( tids[i] & BACKBIT ) {
continue;
}
// skip if we hit our limit
if ( table[tid] >= 4 ) {
continue;
}
// skip over tag/word #i
i++;
// no words in links, unless it is a self link
if ( i < NW && (flags[i] & 0x02) ) {
continue;
}
// the start should be here
int32_t start = -1;
// do not go too far
int32_t max = i + 200;
// find the corresponding back tag for it
for ( ; i < NW && i < max ; i++ ) {
// hey we got it, BUT we got no alnum word first
// so the thing was empty, so loop back to subloop
if ( (tids[i] & BACKBITCOMP) == tid &&
(tids[i] & BACKBIT ) &&
start == -1 ) {
goto subloop;
}
// if we hit another title-ish tag, loop back up
if ( (tids[i] & BACKBITCOMP) == TAG_TITLE || (tids[i] & BACKBITCOMP) == TAG_A ) {
// if no alnum text, restart at the top
if ( start == -1 ) {
goto subloop;
}
// otherwise, break out and see if title works
break;
}
// if we hit a breaking tag...
if ( isBreakingTagId ( tids[i] & BACKBITCOMP ) &&
// do not consider <span> tags breaking for
// our purposes. i saw a <h1><span> setup before.
tids[i] != TAG_SPAN ) {
break;
}
// skip if not alnum word
if ( ! words->isAlnum(i) ) {
continue;
}
// if we hit an alnum word, break out
if ( start == -1 ) {
start = i;
}
}
// if no start was found, must have had a 0 score in there
if ( start == -1 ) {
continue;
}
// if we exhausted the doc, we are done
if ( i >= NW ) {
break;
}
// skip if way too big!
if ( i >= max ) {
continue;
}
// if was too long do not consider a title
if ( i - start > 300 ) {
continue;
}
// . skip if too many bytes
// . this does not include the length of word #i, but #(i-1)
if ( words->getStringSize ( start , i ) > 1000 ) {
continue;
}
// when using pdftohtml, the title tag is the filename when PDF property does not have title tag
if ( tid == TAG_TITLE && contentType == CT_PDF ) {
// skip if title == '/in.[0-9]*'
char* title_start = words->getWord(start);
char* title_end = words->getWord(i);
size_t title_size = title_end - title_start;
const char* result = strnstr( title_start, "/in.", title_size );
if (result != NULL) {
char* endp = NULL;
// do some further verification to avoid screwing up title
if ((strtoll(result + 4, &endp, 10) > 0) && (endp == title_end)) {
continue;
}
}
}
// count it
table[tid]++;
// max it out if we are positive scoring. stop after the
// first positive scoring guy in a section. this might
// hurt the "Hamlet" thing though...
// store a point to the title tag guy. Msg20.cpp needs this
// because the zak's proximity algo uses it in Summary.cpp
// and in Msg20.cpp
// only get the first one! often the 2nd on is in an iframe!! which we now expand into here.
if ( tid == TAG_TITLE && m_titleTagStart == -1 ) {
m_titleTagStart = start;
m_titleTagEnd = i;
// save the candidate # because we always use this
// as the title if we are a root
if ( tti < 0 ) {
tti = n;
}
}
// point to words class of the body that was passed in to us
cptrs[n] = words;
as[n] = start;
bs[n] = i;
if ( tid == TAG_B ) {
types[n] = TT_BOLDTAG;
scores[n] = 1.0;
} else if ( tid == TAG_H1 ) {
types[n] = TT_HTAG;
scores[n] = 1.8;
} else if ( tid == TAG_H2 ) {
types[n] = TT_HTAG;
scores[n] = 1.7;
} else if ( tid == TAG_H3 ) {
types[n] = TT_HTAG;
scores[n] = 1.6;
} else if ( tid == TAG_TITLE ) {
types[n] = TT_TITLETAG;
scores[n] = 3.0;
} else if ( tid == TAG_DIV ) {
types[n] = TT_DIVTAG;
scores[n] = 1.0;
} else if ( tid == TAG_TD ) {
types[n] = TT_TDTAG;
scores[n] = 1.0;
} else if ( tid == TAG_P ) {
types[n] = TT_PTAG;
scores[n] = 1.0;
} else if ( tid == TAG_FONT ) {
types[n] = TT_FONTTAG;
scores[n] = 1.0;
} else if ( tid == TAG_A ) {
types[n] = TT_ATAG;
// . self link is very powerful BUT
// http://www.npr.org/templates/story/story.php?storyId=5417137
// doesn't use it right! so use
// 1.3 instead of 3.0. that has an "onClick" thing in the
// <a> tag, so check for that!
// this was bad for
// http://www.spiritualwoman.net/?cat=191
// so i am demoting from 3.0 to 1.5
scores[n] = 1.5;
}
// count it
n++;
// start loop over at tag #i, for loop does an i++, so negate
// that so this will work
i--;
// break out if too many already. save some for below.
if ( n + 10 >= MAX_TIT_CANDIDATES ) {
break;
}
}
//logf(LOG_DEBUG,"title: took3=%" PRId64,gettimeofdayInMilliseconds()-x);
//x = gettimeofdayInMilliseconds();
// to handle text documents, throw in the first line of text
// as a title candidate, just make the score really low
bool textDoc = (contentType == CT_UNKNOWN || contentType == CT_TEXT);
if (textDoc) {
// make "i" point to first alphabetical word in the document
int32_t i ;
for ( i = 0 ; i < NW && !words->isAlpha(i) ; i++);
// if we got a first alphabetical word, then assume that to be the start of our title
if ( i < NW && n < MAX_TIT_CANDIDATES ) {
// first word in title is "t0"
int32_t t0 = i;
// find end of first line
int32_t numWords = 0;
// set i to the end now. we MUST find a \n to terminate the
// title, otherwise we will not have a valid title
while (i < NW && numWords < maxTitleWords && (words->isAlnum(i) || !words->hasChar(i, '\n'))) {
if(words->isAlnum(i)) {
numWords++;
}
++i;
}
// "t1" is the end
int32_t t1 = -1;
// we must have found our \n in order to set "t1"
if (i <= NW && numWords < maxTitleWords ) {
t1 = i;
}
// set the ptrs
cptrs [n] = words;
// this is the last resort i guess...
scores [n] = 0.5;
types [n] = TT_FIRSTLINE;
as [n] = t0;
bs [n] = t1;
// add it as a candidate if t0 and t1 were valid
if (t0 >= 0 && t1 > t0) {
n++;
}
}
}
//logf(LOG_DEBUG,"title: took4=%" PRId64,gettimeofdayInMilliseconds()-x);
//x = gettimeofdayInMilliseconds();
{
// now add the last url path to contain underscores or hyphens
char *pstart = firstUrl->getPath();
// get first url
Url *fu = firstUrl;
// start at the end
char *p = fu->getUrl() + fu->getUrlLen();
// end pointer
char *pend = NULL;
// come up here for each path component
while ( p >= pstart ) {
// save end
pend = p;
// skip over /
if ( *p == '/' ) {
p--;
}
// now go back to next /
int32_t count = 0;
for ( ; p >= pstart && *p !='/' ; p-- ) {
if ( *p == '_' || *p == '-' ) {
count++;
}
}
// did we get it?
if ( count > 0 ) {
break;
}
}
// did we get any?
if ( p > pstart && n < MAX_TIT_CANDIDATES ) {
// now set words to that
if ( ! tw[ti].set ( p, (pend - p), true, 0 )) {
return false;
}
// point to that
cptrs [n] = &tw[ti];
as [n] = 0;
bs [n] = tw[ti].getNumWords();
scores [n] = 1.0;
types [n] = TT_URLPATH;
// increment since we are using it
ti++;
// advance
n++;
}
}
// save old n
int32_t oldn = n;
// . do not split titles if we are a root url maps.yahoo.com was getting "Maps" for the title
if ( firstUrl->isRoot() ) {
oldn = -2;
}
// point to list of \0 separated titles
const char *rootTitleBuf = NULL;
const char *rootTitleBufEnd = NULL;
// get the root title if we are not root!
if (filteredRootTitleBuf) {
#ifdef _VALGRIND_
VALGRIND_CHECK_MEM_IS_DEFINED(filteredRootTitleBuf,filteredRootTitleBufSize);
#endif
// point to list of \0 separated titles
rootTitleBuf = filteredRootTitleBuf;
rootTitleBufEnd = filteredRootTitleBuf + filteredRootTitleBufSize;
}
{
Matches m;
if ( rootTitleBuf && query ) {
m.setQuery ( query );
}
// convert into an array
int32_t nr = 0;
const char *pr = rootTitleBuf;
const char *rootTitles[20];
int32_t rootTitleLens[20];
// loop over each root title segment
for ( ; pr && pr < rootTitleBufEnd ; pr += strnlen(pr,rootTitleBufEnd-pr) + 1 ) {
// if we had a query...
if ( query ) {
// reset it
m.reset();
// see if root title segment has query terms in it
m.addMatches ( const_cast<char*>(pr), strnlen(pr,rootTitleBufEnd-pr), MF_TITLEGEN, m_niceness );
// if matches query, do NOT add it, we only add it for
// removing from the title of the page...
if ( m.getNumMatches() ) {
continue;
}
}
// point to it. it should start with an alnum already
// since it is the "filtered" list of root titles...
// if not, fix it in xmldoc then.
rootTitles [nr] = pr;
rootTitleLens[nr] = gbstrlen(pr);
// advance
nr++;
// no breaching
if ( nr >= 20 ) break;
}
// now split up candidates in children candidates by tokenizing
// using :, | and - as delimters.
// the hyphen must have a space on at least one side, so "cd-rom" does
// not create a pair of tokens...
// FIX: for the title:
// Best Careers 2009: Librarian - US News and World Report
// we need to recognize "Best Careers 2009: Librarian" as a subtitle
// otherwise we don't get it as the title. so my question is are we
// going to have to do all the permutations at some point? for now
// let's just add in pairs...
for ( int32_t i = 0 ; i < oldn && n + 3 < MAX_TIT_CANDIDATES ; i++ ) {
// stop if no root title segments
if ( nr <= 0 ) break;
// get the word info
Words *w = cptrs[i];
int32_t a = as[i];
int32_t b = bs[i];
// init
int32_t lasta = a;
char prev = false;
// char length in bytes
//int32_t charlen = 1;
// see how many we add
int32_t added = 0;
char *skipTo = NULL;
bool qualified = true;
// . scan the words looking for a token
// . sometimes the candidates end in ": " so put in "k < b-1"
// . made this from k<b-1 to k<b to fix
// "Hot Tub Time Machine (2010) - IMDb" to strip IMDb
for ( int32_t k = a ; k < b && n + 3 < MAX_TIT_CANDIDATES; k++){
// get word
char *wp = w->getWord(k);
// skip if not alnum
if ( ! w->isAlnum(k) ) {
// in order for next alnum word to
// qualify for "clipping" if it matches
// the root title, there has to be more
// than just spaces here, some punct.
// otherwise title
// "T. D. Jakes: Biography from Answers.com"
// becomes
// "T. D. Jakes: Biography from"
qualified=isWordQualified(wp,w->getWordLen(k));
continue;
}
// gotta be qualified!
if ( ! qualified ) continue;
// skip if in root title
if ( skipTo && wp < skipTo ) continue;
// does this match any root page title segments?
int32_t j;
for ( j = 0 ; j < nr ; j++ ) {
// . compare to root title
// . break out if we matched!
if ( ! strncmp( wp, rootTitles[j], rootTitleLens[j] ) ) {
break;
}
}
// if we did not match a root title segment,
// keep on chugging
if ( j >= nr ) continue;
// . we got a root title match!
// . skip over
skipTo = wp + rootTitleLens[j];
// must land on qualified punct then!!
int32_t e = k+1;
for ( ; e<b && w->getWord(e)<skipTo ; e++ );
// ok, word #e must be a qualified punct
if ( e<b &&
! isWordQualified(w->getWord(e),w->getWordLen(e)))
// assume no match then!!
continue;
// if we had a previous guy, reset the end of the
// previous candidate
if ( prev ) {
bs[n-2] = k;
bs[n-1] = k;
}
// . ok, we got two more candidates
// . well, only one more if this is not the 1st time
if ( ! prev ) {
cptrs [n] = cptrs [i];
scores [n] = scores [i];
types [n] = types [i];
as [n] = lasta;
bs [n] = k;
parent [n] = i;
n++;
added++;
}
// the 2nd one
cptrs [n] = cptrs [i];
scores [n] = scores [i];
types [n] = types [i];
as [n] = e + 1;
bs [n] = bs [i];
parent [n] = i;
n++;
added++;
// now add in the last pair as a whole token
cptrs [n] = cptrs [i];
scores [n] = scores [i];
types [n] = types [i];
as [n] = lasta;
bs [n] = bs [i];
parent [n] = i;
n++;
added++;
// nuke the current candidate then since it got
// split up to not contain the root title...
//cptrs[i] = NULL;
// update this
lasta = k+1;
// if we encounter another delimeter we will have to revise bs[n-1], so note that
prev = true;
}
// nuke the current candidate then since it got
// split up to not contain the root title...
if ( added ) {
scores[i] = 0.001;
//cptrs[i] = NULL;
}
// erase the pair if that there was only one token
if ( added == 3 ) n--;
}
}
for ( int32_t i = 0 ; i < n ; i++ ) baseScore[i] = scores[i];
//
// . now punish by 0.85 for every lower case non-stop word it has
// . reward by 1.1 if has a non-stopword in the query
//
for ( int32_t i = 0 ; i < n ; i++ ) {
// point to the words
Words *w = cptrs[i];
// skip if got nuked above
if ( ! w ) {
continue;
}
// the word ptrs
char **wptrs = w->getWordPtrs();
// skip if empty
if ( w->getNumWords() <= 0 ) {
continue;
}
// get the word boundaries
int32_t a = as[i];
int32_t b = bs[i];
// record the boosts
float ncb = 1.0;
float qtb = 1.0;
// a flag
char uncapped = false;
// scan the words in this title candidate
for ( int32_t j = a ; j < b ; j++ ) {
// skip stop words
if ( w->isQueryStopWord( j, langId ) ) {
continue;
}
// punish if uncapitalized non-stopword
if ( ! w->isCapitalized(j) ) {
uncapped = true;
}
// skip if no query
if ( ! query ) {
continue;
}
int64_t wid = w->getWordId(j);
// reward if in the query
if ( query->getWordNum(wid) >= 0 ) {
qtb *= 1.5;
scores[i] *= 1.5;
}
}
// . only punish once if missing a capitalized word hurts us for:
// http://content-uk.cricinfo.com/ausvrsa2008_09/engine/current/match/351682.html
if ( uncapped ) {
ncb *= 1.00;
scores[i] *= 1.00;
}
// punish if a http:// title thingy
char *s = wptrs[a];
int32_t size = w->getStringSize(a,b);
if ( size > 9 && memcmp("http://", s, 7) == 0 ) {
ncb *= .10;
}
if ( size > 14 && memcmp("h\0t\0t\0p\0:\0/\0/", s, 14) == 0 ) {
ncb *= .10;
}
// set these guys
scores[i] *= ncb;
noCapsBoost[i] = ncb;
qtermsBoost[i] = qtb;
}
// . now compare each candidate to the other candidates
// . give a boost if matches
for ( int32_t i = 0 ; i < n ; i++ ) {
// point to the words
Words *w1 = cptrs[i];
// skip if got nuked above
if ( ! w1 ) {
continue;
}
int32_t a1 = as[i];
int32_t b1 = bs[i];
// reset some flags
char localFlag1 = 0;
char localFlag2 = 0;
// record the boost
float iccb = 1.0;
// total boost
float total = 1.0;
// to each other candidate
for ( int32_t j = 0 ; j < n ; j++ ) {
// not to ourselves
if ( j == i ) {
continue;
}
// or our derivatives
if ( parent[j] == i ) {
continue;
}
// or derivates to their parent
if ( parent[i] == j ) {
continue;
}
// only check parents now. do not check kids.
// this was only for when doing percent contained
// not getSimilarity() per se
//if ( parent[j] != -1 ) continue;
// TODO: do not accumulate boosts from a parent
// and its kids, subtitles...
//
// do not compare type X to type Y
if ( types[i] == TT_TITLETAG ) {
if ( types[j] == TT_TITLETAG ) {
continue;
}
}
// do not compare a div candidate to another div cand
// http://friendfeed.com/foxiewire?start=30
// likewise, a TD to another TD
// http://content-uk.cricinfo.com/ausvrsa2008_09/engine/match/351681.html
// ... etc.
if ( types[i] == TT_BOLDTAG ||
types[i] == TT_HTAG ||
types[i] == TT_DIVTAG ||
types[i] == TT_TDTAG ||
types[i] == TT_FONTTAG ) {
if ( types[j] == types[i] ) continue;
}
// . do not compare one kid to another kid
// . i.e. if we got "x | y" as a title and "x | z"
// as a link text, it will emphasize "x" too much
// http://content-uk.cricinfo.com/ausvrsa2008_09/engine/current/match/351682.html
if ( parent[j] != -1 && parent[i] != -1 ) continue;
// . body type tags are mostly mutually exclusive
// . for the legacy.com url mentioned below, we have
// good stuff in <td> tags, so this hurts us...
// . but for the sake of
// http://larvatusprodeo.net/2009/01/07/partisanship-politics-and-participation/
// i put bold tags back
if ( types[i] == TT_LINKTEXTLOCAL ) {
if ( types[j] == TT_LINKTEXTLOCAL ) continue;
}
if ( types[i] == TT_RSSITEMLOCAL ) {
if ( types[j] == TT_RSSITEMLOCAL ) continue;
}
// only compare to one local link text for each i
if ( types[j] == TT_LINKTEXTLOCAL && localFlag1 ) {
continue;
}
if ( types[j] == TT_RSSITEMLOCAL && localFlag2 ) {
continue;
}
if ( types[j] == TT_LINKTEXTLOCAL ) {
localFlag1 = 1;
}
if ( types[j] == TT_RSSITEMLOCAL ) {
localFlag2 = 1;
}
// not link title attr to link title attr either
// fixes http://www.spiritualwoman.net/?cat=191
if ( types[i] == TT_TITLEATT &&
types[j] == TT_TITLEATT )
continue;
// get our words
Words *w2 = cptrs[j];
// skip if got nuked above
if ( ! w2 ) continue;
int32_t a2 = as [j];
int32_t b2 = bs [j];
// how similar is title #i to title #j ?
float fp = getSimilarity ( w2 , a2 , b2 , w1 , a1 , b1 );
// error?
if ( fp == -1.0 ) return false;
// custom boosting...
float boost = 1.0;
if ( fp >= .95 ) boost = 3.0;
else if ( fp >= .90 ) boost = 2.0;
else if ( fp >= .85 ) boost = 1.5;
else if ( fp >= .80 ) boost = 1.4;
else if ( fp >= .75 ) boost = 1.3;
else if ( fp >= .70 ) boost = 1.2;
else if ( fp >= .60 ) boost = 1.1;
else if ( fp >= .50 ) boost = 1.08;
else if ( fp >= .40 ) boost = 1.04;
// limit total
total *= boost;
if ( total > 100.0 ) break;
// if you are matching the url path, that is pretty
// good so give more!
// actually, that would hurt:
// http://michellemalkin.com/2008/12/29/gag-worthy/
// custom boosting!
if ( fp > 0.0 && g_conf.m_logDebugTitle )
logf(LOG_DEBUG,"title: i=%" PRId32" j=%" PRId32" fp=%.02f "
"b=%.02f", i,j,fp,boost);
// apply it
scores[i] *= boost;
iccb *= boost;
}
inCommonCandBoost[i] = iccb;
}
//logf(LOG_DEBUG,"title: took7=%" PRId64,gettimeofdayInMilliseconds()-x);
//x = gettimeofdayInMilliseconds();
// loop over all n candidates
for ( int32_t i = 0 ; i < n ; i++ ) {
// skip if not in the document body
if ( cptrs[i] != words ) continue;
// point to the words
int32_t a1 = as [i];
int32_t b1 = bs [i];
// . loop through this candidates words
// . TODO: use memset here?
for ( int32_t j = a1 ; j <= b1 && j < NW ; j++ ) {
// flag it
flags[j] |= 0x01;
}
}
// free our stuff
if ( flags!=localBuf ) {
mfree (flags, need, "TITLEflags");
}
// now get the highest scoring candidate title
float max = -1.0;
int32_t winner = -1;
for ( int32_t i = 0 ; i < n ; i++ ) {
// skip if got nuked
if ( ! cptrs[i] ) {
continue;
}
if ( winner != -1 && scores[i] <= max ) {
continue;
}
// url path's cannot be titles in and of themselves
if ( types[i] == TT_URLPATH ) {
continue;
}
// skip if empty basically, like if title was exact
// copy of root, then the whole thing got nuked and
// some empty string added, where a > b
if ( as[i] >= bs[i] ) {
continue;
}
// got one
max = scores[i];
// save it
winner = i;
}
// if we are a root, always pick the title tag as the title
if ( oldn == -2 && tti >= 0 ) {
winner = tti;
}
// if no winner, all done. no title
if ( winner == -1 ) {
// last resort use file name
if ((contentType == CT_PDF) && (firstUrl->getFilenameLen() != 0)) {
Words w;
w.set(firstUrl->getFilename(), firstUrl->getFilenameLen(), true);
if (!copyTitle(&w, 0, w.getNumWords())) {
return false;
}
}
return true;
}
// point to the words class of the winner
Words *w = cptrs[winner];
// skip if got nuked above
if ( ! w ) { char *xx=NULL;*xx=0; }
// need to make our own Pos class if title not from body
Pos tp;
if ( w != words ) {
// set "Scores" ptr to NULL. we assume all are positive scores
if ( ! tp.set ( w ) ) {
return false;
}
}
// the string ranges from word #a up to and including word #b
int32_t a = as[winner];
int32_t b = bs[winner];
// sanity check
if ( a < 0 || b > w->getNumWords() ) { char*xx=NULL;*xx=0; }
// save the title
if ( ! copyTitle(w, a, b) ) {
return false;
}
/*
// debug logging
SafeBuf sb;
SafeBuf *pbuf = &sb;
log("title: candidates for %s",xd->getFirstUrl()->getUrl() );
pbuf->safePrintf("<div stype=\"border:1px solid black\">");
pbuf->safePrintf("<b>***Finding Title***</b><br>\n");
pbuf->safePrintf("<table cellpadding=5 border=2><tr>"
"<td colspan=20><center><b>Title Generation</b>"
"</center></td>"
"</tr>\n<tr>"
"<td>#</td>"
"<td>type</td>"
"<td>parent</td>"
"<td>base score</td>"
"<td>format penalty</td>"
"<td>query term boost</td>"
"<td>candidate intersection boost</td>"
"<td>FINAL SCORE</td>"
"<td>title</td>"
"</tr>\n" );
// print out all candidates
for ( int32_t i = 0 ; i < n ; i++ ) {
char *ts = "unknown";
if ( types[i] == TT_LINKTEXTLOCAL ) ts = "local inlink text";
if ( types[i] == TT_LINKTEXTREMOTE ) ts = "remote inlink text";
if ( types[i] == TT_RSSITEMLOCAL ) ts = "local rss title";
if ( types[i] == TT_RSSITEMREMOTE ) ts = "remote rss title";
if ( types[i] == TT_BOLDTAG ) ts = "bold tag";
if ( types[i] == TT_HTAG ) ts = "header tag";
if ( types[i] == TT_TITLETAG ) ts = "title tag";
if ( types[i] == TT_FIRSTLINE ) ts = "first line in text";
if ( types[i] == TT_FONTTAG ) ts = "font tag";
if ( types[i] == TT_ATAG ) ts = "anchor tag";
if ( types[i] == TT_DIVTAG ) ts = "div tag";
if ( types[i] == TT_TDTAG ) ts = "td tag";
if ( types[i] == TT_PTAG ) ts = "p tag";
if ( types[i] == TT_URLPATH ) ts = "url path";
if ( types[i] == TT_TITLEATT ) ts = "title attribute";
// get the title
pbuf->safePrintf(
"<tr>"
"<td>#%" PRId32"</td>"
"<td><nobr>%s</nobr></td>"
"<td>%" PRId32"</td>"
"<td>%0.2f</td>" // baseScore
"<td>%0.2f</td>"
"<td>%0.2f</td>"
"<td>%0.2f</td>"
"<td>%0.2f</td>"
"<td>",
i,
ts ,
parent[i],
baseScore[i],
noCapsBoost[i],
qtermsBoost[i],
inCommonCandBoost[i],
scores[i]);
// ptrs
Words *w = cptrs[i];
int32_t a = as[i];
int32_t b = bs[i];
// skip if no words
if ( w->getNumWords() <= 0 ) continue;
// the word ptrs
char **wptrs = w->getWordPtrs();
// string ptrs
char *ptr = wptrs[a];//w->getWord(a);
int32_t size = w->getStringSize(a,b);
// it is utf8
pbuf->safeMemcpy ( ptr , size );
// end the line
pbuf->safePrintf("</td></tr>\n");
}
pbuf->safePrintf("</table>\n<br>\n");
// log these for now
log("title: %s",sb.getBufStart());
*/
return true;
}
// . return length stored into "buf"
// . content must be NULL terminated
// . if "useAnchors" is true we do click and scroll
// . if "isQueryTerms" is true, we do typical anchors in a special way
int32_t Highlight::set ( SafeBuf *sb,
//char *buf ,
//int32_t bufLen ,
char *content ,
int32_t contentLen ,
// primary language of the document (for synonyms)
char docLangId ,
Query *q ,
bool doStemming ,
bool useAnchors ,
const char *baseUrl ,
const char *frontTag ,
const char *backTag ,
int32_t fieldCode ,
int32_t niceness ) {
Words words;
if ( ! words.set ( content ,
contentLen ,
TITLEREC_CURRENT_VERSION,
true , // computeId
true ) ) // has html entites?
return -1;
int32_t version = TITLEREC_CURRENT_VERSION;
Bits bits;
if ( ! bits.set (&words,version,niceness) ) return -1;
Phrases phrases;
if ( !phrases.set(&words,&bits,true,false,version,niceness))return -1;
//SafeBuf langBuf;
//if ( !setLangVec ( &words , &langBuf , niceness )) return 0;
//uint8_t *langVec = (uint8_t *)langBuf.getBufStart();
// make synonyms
//Synonyms syns;
//if(!syns.set(&words,NULL,docLangId,&phrases,niceness,NULL)) return 0;
Matches matches;
matches.setQuery ( q );
if ( ! matches.addMatches ( &words , &phrases ) ) return -1;
// store
m_numMatches = matches.getNumMatches();
return set ( sb ,
//buf ,
//bufLen ,
&words ,
&matches ,
doStemming ,
useAnchors ,
baseUrl ,
frontTag ,
backTag ,
fieldCode ,
q );
}