require 'rubygems' require 'nokogiri' require 'guess_html_encoding' module Readability class Document DEFAULT_OPTIONS = { :retry_length => 250, :min_text_length => 25, :remove_unlikely_candidates => true, :weight_classes => true, :clean_conditionally => true, :remove_empty_nodes => true, :min_image_width => 130, :min_image_height => 80, :ignore_image_format => [] }.freeze attr_accessor :options, :html, :best_candidate, :candidates, :best_candidate_has_image def initialize(input, options = {}) @options = DEFAULT_OPTIONS.merge(options) @input = input if RUBY_VERSION =~ /^1\.9\./ && !@options[:encoding] @input = GuessHtmlEncoding.encode(@input, @options[:html_headers]) unless @options[:do_not_guess_encoding] @options[:encoding] = @input.encoding.to_s end @input = @input.gsub(REGEXES[:replaceBrsRe], '

').gsub(REGEXES[:replaceFontsRe], '<\1span>') @remove_unlikely_candidates = @options[:remove_unlikely_candidates] @weight_classes = @options[:weight_classes] @clean_conditionally = @options[:clean_conditionally] @best_candidate_has_image = true make_html end def prepare_candidates @html.css("script, style").each { |i| i.remove } remove_unlikely_candidates! if @remove_unlikely_candidates transform_misused_divs_into_paragraphs! @candidates = score_paragraphs(options[:min_text_length]) @best_candidate = select_best_candidate(@candidates) end def make_html @html = Nokogiri::HTML(@input, nil, @options[:encoding]) # In case document has no body, such as from empty string or redirect @html = Nokogiri::HTML('', nil, @options[:encoding]) if @html.css('body').length == 0 # Remove html comment tags @html.xpath('//comment()').each { |i| i.remove } end def images(content=nil, reload=false) begin require 'mini_magick' rescue LoadError raise "Please install mini_magick in order to use the #images feature." end @best_candidate_has_image = false if reload prepare_candidates list_images = [] tested_images = [] content = @best_candidate[:elem] unless reload return list_images if content.nil? elements = content.css("img").map(&:attributes) elements.each do |element| next unless element["src"] url = element["src"].value height = element["height"].nil? ? 0 : element["height"].value.to_i width = element["width"].nil? ? 0 : element["width"].value.to_i format = File.extname(url).gsub(".", "") image = {:width => width, :height => height, :format => format} image = load_image(url) if url =~ /\Ahttps?:\/\//i && (height.zero? || width.zero?) next unless image if tested_images.include?(url) debug("Image was tested: #{url}") next end tested_images.push(url) if image_meets_criteria?(image) list_images << url else debug("Image discarded: #{url} - height: #{image[:height]} - width: #{image[:width]} - format: #{image[:format]}") end end (list_images.empty? and content != @html) ? images(@html, true) : list_images end def load_image(url) begin MiniMagick::Image.open(url) rescue => e debug("Image error: #{e}") nil end end def image_meets_criteria?(image) return false if options[:ignore_image_format].include?(image[:format].downcase) image[:width] >= (options[:min_image_width] || 0) && image[:height] >= (options[:min_image_height] || 0) end REGEXES = { :unlikelyCandidatesRe => /combx|comment|community|disqus|extra|foot|header|menu|remark|rss|shoutbox|sidebar|sponsor|ad-break|agegate|pagination|pager|popup/i, :okMaybeItsACandidateRe => /and|article|body|column|main|shadow/i, :positiveRe => /article|body|content|entry|hentry|main|page|pagination|post|text|blog|story/i, :negativeRe => /combx|comment|com-|contact|foot|footer|footnote|masthead|media|meta|outbrain|promo|related|scroll|shoutbox|sidebar|sponsor|shopping|tags|tool|widget/i, :divToPElementsRe => /<(a|blockquote|dl|div|img|ol|p|pre|table|ul)/i, :replaceBrsRe => /(]*>[ \n\r\t]*){2,}/i, :replaceFontsRe => /<(\/?)font[^>]*>/i, :trimRe => /^\s+|\s+$/, :normalizeRe => /\s{2,}/, :killBreaksRe => /((\s| ?)*){1,}/, :videoRe => /http:\/\/(www\.)?(youtube|vimeo)\.com/i } def title title = @html.css("title").first title ? title.text : nil end # Look through the @html document looking for the author # Precedence Information here on the wiki: (TODO attach wiki URL if it is accepted) # Returns nil if no author is detected def author # Let's grab this author: # author_elements = @html.xpath('//meta[@name = "dc.creator"]') unless author_elements.empty? author_elements.each do |element| if element['content'] return element['content'].strip end end end # Now let's try to grab this # ByAustin Fonacier #

By
Austin Fonacier
author_elements = @html.xpath('//*[contains(@class, "vcard")]//*[contains(@class, "fn")]') unless author_elements.empty? author_elements.each do |element| if element.text return element.text.strip end end end # Now let's try to grab this # # TODO: strip out the (rel)? author_elements = @html.xpath('//a[@rel = "author"]') unless author_elements.empty? author_elements.each do |element| if element.text return element.text.strip end end end author_elements = @html.xpath('//*[@id = "author"]') unless author_elements.empty? author_elements.each do |element| if element.text return element.text.strip end end end end def content(remove_unlikely_candidates = :default) @remove_unlikely_candidates = false if remove_unlikely_candidates == false prepare_candidates article = get_article(@candidates, @best_candidate) cleaned_article = sanitize(article, @candidates, options) if article.text.strip.length < options[:retry_length] if @remove_unlikely_candidates @remove_unlikely_candidates = false elsif @weight_classes @weight_classes = false elsif @clean_conditionally @clean_conditionally = false else # nothing we can do return cleaned_article end make_html content else cleaned_article end end def get_article(candidates, best_candidate) # Now that we have the top candidate, look through its siblings for content that might also be related. # Things like preambles, content split by ads that we removed, etc. sibling_score_threshold = [10, best_candidate[:content_score] * 0.2].max output = Nokogiri::XML::Node.new('div', @html) best_candidate[:elem].parent.children.each do |sibling| append = false append = true if sibling == best_candidate[:elem] append = true if candidates[sibling] && candidates[sibling][:content_score] >= sibling_score_threshold if sibling.name.downcase == "p" link_density = get_link_density(sibling) node_content = sibling.text node_length = node_content.length if node_length > 80 && link_density < 0.25 append = true elsif node_length < 80 && link_density == 0 && node_content =~ /\.( |$)/ append = true end end if append sibling_dup = sibling.dup # otherwise the state of the document in processing will change, thus creating side effects sibling_dup.name = "div" unless %w[div p].include?(sibling.name.downcase) output << sibling_dup end end output end def select_best_candidate(candidates) sorted_candidates = candidates.values.sort { |a, b| b[:content_score] <=> a[:content_score] } debug("Top 5 candidates:") sorted_candidates[0...5].each do |candidate| debug("Candidate #{candidate[:elem].name}##{candidate[:elem][:id]}.#{candidate[:elem][:class]} with score #{candidate[:content_score]}") end best_candidate = sorted_candidates.first || { :elem => @html.css("body").first, :content_score => 0 } debug("Best candidate #{best_candidate[:elem].name}##{best_candidate[:elem][:id]}.#{best_candidate[:elem][:class]} with score #{best_candidate[:content_score]}") best_candidate end def get_link_density(elem) link_length = elem.css("a").map(&:text).join("").length text_length = elem.text.length link_length / text_length.to_f end def score_paragraphs(min_text_length) candidates = {} @html.css("p,td").each do |elem| parent_node = elem.parent grand_parent_node = parent_node.respond_to?(:parent) ? parent_node.parent : nil inner_text = elem.text # If this paragraph is less than 25 characters, don't even count it. next if inner_text.length < min_text_length candidates[parent_node] ||= score_node(parent_node) candidates[grand_parent_node] ||= score_node(grand_parent_node) if grand_parent_node content_score = 1 content_score += inner_text.split(',').length content_score += [(inner_text.length / 100).to_i, 3].min candidates[parent_node][:content_score] += content_score candidates[grand_parent_node][:content_score] += content_score / 2.0 if grand_parent_node end # Scale the final candidates score based on link density. Good content should have a # relatively small link density (5% or less) and be mostly unaffected by this operation. candidates.each do |elem, candidate| candidate[:content_score] = candidate[:content_score] * (1 - get_link_density(elem)) end candidates end def class_weight(e) weight = 0 return weight unless @weight_classes if e[:class] && e[:class] != "" if e[:class] =~ REGEXES[:negativeRe] weight -= 25 end if e[:class] =~ REGEXES[:positiveRe] weight += 25 end end if e[:id] && e[:id] != "" if e[:id] =~ REGEXES[:negativeRe] weight -= 25 end if e[:id] =~ REGEXES[:positiveRe] weight += 25 end end weight end def score_node(elem) content_score = class_weight(elem) case elem.name.downcase when "div" content_score += 5 when "blockquote" content_score += 3 when "form" content_score -= 3 when "th" content_score -= 5 end { :content_score => content_score, :elem => elem } end def debug(str) puts str if options[:debug] end def remove_unlikely_candidates! @html.css("*").each do |elem| str = "#{elem[:class]}#{elem[:id]}" if str =~ REGEXES[:unlikelyCandidatesRe] && str !~ REGEXES[:okMaybeItsACandidateRe] && (elem.name.downcase != 'html') && (elem.name.downcase != 'body') debug("Removing unlikely candidate - #{str}") elem.remove end end end def transform_misused_divs_into_paragraphs! @html.css("*").each do |elem| if elem.name.downcase == "div" # transform
s that do not contain other block elements into

s if elem.inner_html !~ REGEXES[:divToPElementsRe] debug("Altering div(##{elem[:id]}.#{elem[:class]}) to p"); elem.name = "p" end else # wrap text nodes in p tags # elem.children.each do |child| # if child.text? # debug("wrapping text node with a p") # child.swap("

#{child.text}

") # end # end end end end def sanitize(node, candidates, options = {}) node.css("h1, h2, h3, h4, h5, h6").each do |header| header.remove if class_weight(header) < 0 || get_link_density(header) > 0.33 end node.css("form, object, iframe, embed").each do |elem| elem.remove end if @options[:remove_empty_nodes] # remove

tags that have no text content - this will also remove p tags that contain only images. node.css("p").each do |elem| elem.remove if elem.content.strip.empty? end end # Conditionally clean s,