{ "translatorID":"b6d0a7a-d076-48ae-b2f0-b6de28b194e", "translatorType":4, "label":"ScienceDirect", "creator":"Michael Berkowitz", "target":"https?://[^/]*science-?direct\\.com[^/]*/science(\\/article)?(\\?(?:.+\\&|)ob=(?:ArticleURL|ArticleListURL|PublicationURL))?", "minVersion":"1.0.0b3.r1", "maxVersion":null, "priority":100, "inRepository":true, "lastUpdated":"2009-10-31 20:35:00" } function detectWeb(doc, url) { if ((url.indexOf("_ob=DownloadURL") != -1) || doc.title == "ScienceDirect Login" || doc.title == "ScienceDirect - Dummy") { return false; } if((!url.match("pdf") && url.indexOf("_ob=ArticleURL") == -1 && url.indexOf("/article/") == -1) || url.indexOf("/journal/") != -1) { return "multiple"; } else if (!url.match("pdf")) { return "journalArticle"; } } function doWeb(doc, url) { var namespace = doc.documentElement.namespaceURI; var nsResolver = namespace ? function(prefix) { if (prefix == 'x') return namespace; else return null; } : null; if (doc.evaluate('//*[contains(@src, "exportarticle_a.gif")]', doc, nsResolver, XPathResult.ANY_TYPE, null).iterateNext()) { var articles = new Array(); if(detectWeb(doc, url) == "multiple") { //search page var items = new Object(); var xpath; if (url.indexOf("_ob=PublicationURL") != -1) { // not sure if this case still arises. may need to be fixed at some point xpath = '//table[@class="resultRow"]/tbody/tr/td[2]/a'; } else { xpath = '//div[@class="font3"][@id="bodyMainResults"]/table/tbody/tr/td[2]/a'; } var rows = doc.evaluate(xpath, doc, nsResolver, XPathResult.ANY_TYPE, null); var next_row; while (next_row = rows.iterateNext()) { var title = next_row.textContent; var link = next_row.href; if (!title.match(/PDF \(/) && !title.match(/Related Articles/)) items[link] = title; } items = Zotero.selectItems(items); for (var i in items) { articles.push(i); } var sets = []; for each (article in articles) { sets.push({article:article}); } } else { articles = [url]; var sets =[{currentdoc:doc}]; } if(articles.length == 0) { Zotero.debug('no items'); return; } var scrape = function(newDoc, set) { var PDF; var tempPDF = newDoc.evaluate('//a[@class="noul" and div/div[contains(text(), "PDF")]]', newDoc, nsResolver, XPathResult.ANY_TYPE, null).iterateNext(); if (!tempPDF) { // PDF xpath failed, lets try another tempPDF = newDoc.evaluate('//a[@class="noul" and contains(text(), "PDF")]', newDoc, nsResolver, XPathResult.ANY_TYPE, null).iterateNext(); if (!tempPDF) { // second PDF xpath failed set PDF to null to avoid item.attachments PDF = null; } else { PDF = tempPDF.href; // second xpath succeeded, use that link } } else { PDF = tempPDF.href; // first xpath succeeded, use that link } var url = newDoc.location.href; var get = newDoc.evaluate('//a[img[contains(@src, "exportarticle_a.gif")]]', newDoc, nsResolver, XPathResult.ANY_TYPE, null).iterateNext().href; // if the PDF is available make it an attachment otherwise only use snapshot. var attachments; if (PDF) { attachments = [ {url:url, title:"ScienceDirect Snapshot", mimeType:"text/html"}, {url:PDF, title:"ScienceDirect Full Text PDF", mimeType:"application/pdf"} // Sometimes PDF is null...I hope that is ok ]; } else { attachments = [ {url:url, title:"ScienceDirect Snapshot", mimeType:"text/html"} ]; } // This does not work, not sure why. //var doi = newDoc.evaluate('//a[contains(text(), "doi")]/text()', newDoc, nsResolver, XPathResult.ANY_TYPE, null).iterateNext(); //Zotero.debug(doi); //doi = doi.textContent.substr(4); // pass these values to the next function //set.doi = doi; set.url = url; set.get = get; set.attachments = attachments; return set; }; var first = function(set, next) { var article = set.article; Zotero.Utilities.processDocuments(article, function(doc){ set = scrape(doc, set); next(); }); }; var second = function(set, next) { var url = set.url; var get = set.get; Zotero.Utilities.HTTP.doGet(get, function(text) { var md5 = text.match(/]+)>/)[1]; var acct = text.match(/]+)>/)[1]; var userid = text.match(/]+)>/)[1]; var uoikey = text.match(/]+)>/)[1]; if (text.match(/]+)>/)) { var alid = text.match(/]+)>/)[1]; } if (alid) { var docID = "_ArticleListID=" + alid + "&_uoikey=" + uoikey; } else { var docID = "_uoikey=" + uoikey; } var post = "_ob=DownloadURL&_method=finish&_acct=" + acct + "&_userid=" + userid + "&_docType=FLA&" + docID + "&md5=" + md5 + "&count=1&JAVASCRIPT_ON=Y&format=cite-abs&citation-type=RIS&Export=Export&x=26&y=17"; var baseurl = url.match(/https?:\/\/[^/]+\//)[0]; set.post = post; set.baseurl = baseurl; next(); }); }; var third = function(set, next) { var baseurl = set.baseurl; var post = set.post; var attachments = set.attachments; Zotero.Utilities.HTTP.doPost(baseurl + 'science', post, function(text) { var translator = Zotero.loadTranslator("import"); translator.setTranslator("32d59d2d-b65a-4da4-b0a3-bdd3cfb979e7"); translator.setString(text); translator.setHandler("itemDone", function(obj, item) { item.attachments = attachments; if(item.notes[0]) { item.abstractNote = item.notes[0].note; item.notes = new Array(); } item.DOI = item.DOI.substr(10); item.complete(); }); translator.translate(); next(); }, false, 'windows-1252'); }; if(detectWeb(doc, url) == "journalArticle") { Zotero.debug("Single"); var set = scrape(doc, {}); second(set, function(){ third(set, function(){ Zotero.done(); }); }); } else { var callbacks = [first, second, third]; Zotero.Utilities.processAsync(sets, callbacks, function() {Zotero.done()}); } } else { var sets = []; var articles = new Array(); if (detectWeb(doc, url) == "multiple") { var items = new Object(); if (url.indexOf("_ob=PublicationURL") != -1) { xpath = '//table[@class="resultRow"]/tbody/tr/td[2]/a'; // not sure whether this case still exists } else { xpath = '//div[@class="font3"][@id="bodyMainResults"]/table/tbody/tr/td[2]/a'; } var rows = doc.evaluate(xpath, doc, nsResolver, XPathResult.ANY_TYPE, null); var next_row; while (next_row = rows.iterateNext()) { var title = next_row.textContent; var link = next_row.href; if (!title.match(/PDF \(/) && !title.match(/Related Articles/)) items[link] = title; } items = Zotero.selectItems(items); for (var i in items) { articles.push(i); } } else { articles = [url]; } if(articles.length == 0) { Zotero.debug('no items'); return; } for each (article in articles) { sets.push({article:article}); } var unauthScrape = function(doc2) { var item = new Zotero.Item("journalArticle"); item.repository = "ScienceDirect"; item.url = doc2.location.href; var title = doc2.title.match(/^[^-]+\-([^:]+):(.*)$/); item.title = Zotero.Utilities.trimInternal(title[2]); item.publicationTitle = Zotero.Utilities.trimInternal(title[1]); voliss = doc2.evaluate('//div[@class="pageText"][@id="sdBody"]/table/tbody/tr/td[1]', doc2, nsResolver, XPathResult.ANY_TYPE, null).iterateNext().textContent; if (voliss.match(/Volume\s+\d+/)) item.volume = voliss.match(/Volume\s+(\d+)/)[1]; if (voliss.match(/Issues?\s+[^,]+/)) item.issue = voliss.match(/Issues?\s+([^,]+)/)[1]; if (voliss.match(/(J|F|M|A|S|O|N|D)\w+\s+\d{4}/)) item.date = voliss.match(/(J|F|M|A|S|O|N|D)\w+\s+\d{4}/)[0]; if (voliss.match(/Pages?\s+[^,^\s]+/)) item.pages = voliss.match(/Pages?\s+([^,^\s]+)/)[1]; /* FIXME: pulling full article content var abspath = '//div[@class="articleHeaderInner"][@id="articleHeader"]/div[@class="articleText"]/p'; var absx = doc2.evaluate(abspath, doc2, nsResolver, XPathResult.ANY_TYPE, null); var ab; item.abstractNote = ""; while (ab = absx.iterateNext()) { item.abstractNote += Zotero.Utilities.trimInternal(ab.textContent) + " "; } if (item.abstractNote.substr(0, 7) == "Summary") { item.abstractNote = item.abstractNote.substr(9); } */ var tagpath = '//div[@class="articleText"]/p[strong[starts-with(text(), "Keywords:")]]'; if (doc2.evaluate(tagpath, doc2, nsResolver, XPathResult.ANY_TYPE, null).iterateNext()) { if (doc2.evaluate(tagpath, doc2, nsResolver, XPathResult.ANY_TYPE, null).iterateNext().textContent.split(":")[1]) { var tags = doc2.evaluate(tagpath, doc2, nsResolver, XPathResult.ANY_TYPE, null).iterateNext().textContent.split(":")[1].split(";"); for (var i in tags) { item.tags.push(Zotero.Utilities.trimInternal(tags[i])); } } } /* FIXME: producing empty pages item.attachments.push({url:doc2.location.href, title:"ScienceDirect Snapshot", mimeType:"text/html"}); */ return item; }; var first = function(set, next) { var article = set.article; Zotero.Utilities.processDocuments(article, function(doc2) { set.item = unauthScrape(doc2); next(); }); }; var second = function(set, next) { var item = set.item; Zotero.Utilities.HTTP.doGet(item.url, function(text) { item.DOI = text.match(/>doi:([^<]*)/)[1]; try { var aus = text.match(/\s+

.*<\/strong>/)[0].replace(//g, "$").replace(/<\/sup>/g, "$"); aus = aus.replace(/\$[^$]*\$/g, ""); aus = aus.replace(/]*>/g, "$").replace(/<\/a[^>]*>/g, "$"); aus = aus.replace(/\$[^$]*\$/g, ""); aus = Zotero.Utilities.cleanTags(aus); aus = aus.split(/(,|and)/); for (var a in aus) { if (aus[a] != "," && aus[a] != "and" && aus[a].match(/\w+/)) { item.creators.push(Zotero.Utilities.cleanAuthor(Zotero.Utilities.unescapeHTML(Zotero.Utilities.trimInternal(aus[a]), "author"))); } } } catch(e) { Zotero.debug("No Authors listed."); } item.complete(); next(); }); }; var callbacks = [first, second]; if(detectWeb(doc, url) == "journalArticle") { second({item:unauthScrape(doc)}, function() {Zotero.done()}); } else { Zotero.Utilities.processAsync(sets, callbacks, function() {Zotero.done()}); } } Zotero.wait(); }