1 | package org.greenstone.atea;
|
---|
2 |
|
---|
3 |
|
---|
4 | import java.io.*;
|
---|
5 | import java.util.Properties;
|
---|
6 | import java.util.zip.GZIPInputStream;
|
---|
7 | import java.util.Iterator;
|
---|
8 | import java.util.Set;
|
---|
9 | import java.util.TreeSet;
|
---|
10 |
|
---|
11 | import org.apache.log4j.Logger;
|
---|
12 |
|
---|
13 | /**
|
---|
14 | * The main() method of this class takes a folder of warc.wet(.gz) files and goes through
|
---|
15 | * the WET records in each, putting each WET record into a file. Each file is put into a
|
---|
16 | * keep or discard folder, based on content-length and number of lines.
|
---|
17 | * A single instance of the WETProcessor class processes a single unzipped warc.wet file.
|
---|
18 | *
|
---|
19 | * To compile, including the jars in lib/ for compiling.
|
---|
20 | * maori-lang-detection/src$ javac -cp ".:../lib/*" org/greenstone/atea/WETProcessor.java
|
---|
21 | *
|
---|
22 | * To run, passing the log4j and other properties files in conf/ folder:
|
---|
23 | * maori-lang-detection/src$ java -cp ".:../conf:../lib/*" org.greenstone.atea.WETProcessor <folder containing warc.wet(.gz) files> <outputFolder>
|
---|
24 | *
|
---|
25 | * e.g.
|
---|
26 | * - java -cp ".:../conf:../lib/*" org.greenstone.atea.WETProcessor ../tmp/processWET /Scratch/ak19/gs3-extensions/maori-lang-detection/tmp/processedWET
|
---|
27 | * - java -cp ".:../conf:../lib/*" org.greenstone.atea.WETProcessor ../tmp/processWET /Scratch/ak19/gs3-extensions/maori-lang-detection/tmp/processedWET 2>&1 | less
|
---|
28 | *
|
---|
29 | */
|
---|
30 | public class WETProcessor {
|
---|
31 | private static Logger logger = Logger.getLogger(org.greenstone.atea.WETProcessor.class.getName());
|
---|
32 |
|
---|
33 | // WARC WET header lines and header line prefixes of interest
|
---|
34 | static final String WARC_RECORD_START = "WARC/1.0";
|
---|
35 | static final String WARC_INFO_HEADER = "WARC-Type: warcinfo";
|
---|
36 | static final String WARC_TARGET_URI_HEADER_PREFIX = "WARC-Target-URI:";
|
---|
37 | static final String WARC_CONTENT_LENGTH_HEADER_PREFIX = "Content-Length:";
|
---|
38 |
|
---|
39 | private final String WETFileID;
|
---|
40 | private final File inFile;
|
---|
41 |
|
---|
42 | private int recordCount = 0;
|
---|
43 |
|
---|
44 | /** Handle to a CCWETProcessor that processes a set of WET files
|
---|
45 | * Whereas a WETProcessor instance only processes a single WET file
|
---|
46 | * containing multiple WET records.
|
---|
47 | */
|
---|
48 | private CCWETProcessor batchProcessor;
|
---|
49 |
|
---|
50 | /**
|
---|
51 | * WET processor processes a single warc.wet file containing multiple WET records
|
---|
52 | * containing text identified as primary langcode=mri. Each individual WET record is written
|
---|
53 | * out to a uniquely named file in either the keep or discard folder depending on the WET
|
---|
54 | * record's content length and number of lines of actual content (excluding WARC headers).
|
---|
55 | */
|
---|
56 | public WETProcessor(File inFile, CCWETProcessor batchProcessor) {
|
---|
57 | this.batchProcessor = batchProcessor;
|
---|
58 |
|
---|
59 | this.inFile = inFile;
|
---|
60 | // We just want a unique recordID prefix, which we get from the wet file name suffix:
|
---|
61 | // inFile name looks something like MAORI-CC-2019-30-20190902100139-000000.warc.wet
|
---|
62 | // the prefix will be everything after the last hyphen and without file extension,
|
---|
63 | // so "000000" in our example. Then suffix the recordCount (keeping track of the current
|
---|
64 | // WET record) to get a unique filename to store each WET record into.
|
---|
65 |
|
---|
66 | String fileID = inFile.getName();
|
---|
67 | fileID = fileID.substring(fileID.lastIndexOf("-")+1);
|
---|
68 | fileID = fileID.substring(0, fileID.indexOf("."));
|
---|
69 | this.WETFileID = fileID;
|
---|
70 | }
|
---|
71 |
|
---|
72 | public int processWETFile() {
|
---|
73 | File keepURLsFile = this.batchProcessor.keepURLsFile;
|
---|
74 | File discardURLsFile = this.batchProcessor.discardURLsFile;
|
---|
75 |
|
---|
76 | StringBuilder record = null;
|
---|
77 | String line = null;
|
---|
78 | boolean readingRecord = false;
|
---|
79 |
|
---|
80 | String WARCtargetURI = "";
|
---|
81 |
|
---|
82 | //int recordCount = 0;
|
---|
83 |
|
---|
84 | int contentLength = -1; // of record
|
---|
85 | int lineCount = -1; // actual number of non-empty lines in record body (i.e. excludes WET/WARC headers)
|
---|
86 |
|
---|
87 | // read from WETfile
|
---|
88 | try (
|
---|
89 | BufferedReader reader = new BufferedReader(new FileReader(this.inFile));
|
---|
90 | BufferedWriter keepURLsWriter = new BufferedWriter(new FileWriter(keepURLsFile, true));
|
---|
91 | BufferedWriter discardURLsWriter = new BufferedWriter(new FileWriter(discardURLsFile, true)); // true to append
|
---|
92 | ) {
|
---|
93 |
|
---|
94 | while((line = reader.readLine()) != null) { // readLine removes newline separator
|
---|
95 |
|
---|
96 | if(line.startsWith(WARC_INFO_HEADER)) { // "WARC-Type: warcinfo"
|
---|
97 | readingRecord = false;
|
---|
98 | record = null; // drop this record, which is just an info record not actual web page's text
|
---|
99 | recordCount--;
|
---|
100 | continue;
|
---|
101 | }
|
---|
102 |
|
---|
103 | if(line.startsWith(WARC_RECORD_START)) { // "WARC/1.0" means finished previous WET record
|
---|
104 | // process any previous record
|
---|
105 | if(record != null) {
|
---|
106 | processWETrecord(keepURLsWriter, discardURLsWriter,
|
---|
107 | recordCount, contentLength, lineCount,
|
---|
108 | WARCtargetURI, record.toString());
|
---|
109 | record = null;
|
---|
110 | contentLength = -1;
|
---|
111 | lineCount = -1;
|
---|
112 | }
|
---|
113 |
|
---|
114 | recordCount++;
|
---|
115 | // get ready to start a new record
|
---|
116 | readingRecord = true;
|
---|
117 | record = new StringBuilder();
|
---|
118 | }
|
---|
119 |
|
---|
120 | if(readingRecord) { // append current line to current record
|
---|
121 |
|
---|
122 | if(line.startsWith(WARC_TARGET_URI_HEADER_PREFIX)) { // "WARC-Target-URI:"
|
---|
123 | // get and store the value
|
---|
124 | WARCtargetURI = line.substring(WARC_TARGET_URI_HEADER_PREFIX.length()).trim();
|
---|
125 | }
|
---|
126 |
|
---|
127 | record.append(line + "\n"); // add back (unix style) line ending
|
---|
128 |
|
---|
129 | // if the line is non-empty
|
---|
130 | // AND if we've started counting lines, which happens only when the current
|
---|
131 | // line is past WARC/WET headers and we're into the actual body portion
|
---|
132 | // of the WET record,
|
---|
133 | // start incrementing the line counter.
|
---|
134 | if(lineCount >= 0 && !line.trim().equals("")) {
|
---|
135 | lineCount++;
|
---|
136 | }
|
---|
137 | else if(line.startsWith(WARC_CONTENT_LENGTH_HEADER_PREFIX)) { // "Content-Length:"
|
---|
138 | String contentLengthStr = line.substring(WARC_CONTENT_LENGTH_HEADER_PREFIX.length()).trim();
|
---|
139 | contentLength = Integer.parseInt(contentLengthStr);
|
---|
140 | lineCount = 0;
|
---|
141 | }
|
---|
142 |
|
---|
143 | }
|
---|
144 |
|
---|
145 | }
|
---|
146 |
|
---|
147 | // flush the last record. If it was a warcinfo record, record would be null here
|
---|
148 | if(record != null) {
|
---|
149 | processWETrecord(keepURLsWriter, discardURLsWriter,
|
---|
150 | recordCount, contentLength, lineCount,
|
---|
151 | WARCtargetURI, record.toString());
|
---|
152 | record = null;
|
---|
153 | }
|
---|
154 |
|
---|
155 | } catch(IOException ioe) {
|
---|
156 | ioe.printStackTrace();
|
---|
157 | }
|
---|
158 |
|
---|
159 | return recordCount;
|
---|
160 | }
|
---|
161 |
|
---|
162 | public int getRecordCount() { return this.recordCount; }
|
---|
163 |
|
---|
164 | /**
|
---|
165 | * Determines if a WET record belongs in the keep or discard pile depending on if it
|
---|
166 | * contains enough text, based on contentLength and line count of the record body.
|
---|
167 | * Then writes out the WET record to a uniquely named file in the keep or discard folder,
|
---|
168 | * and writes out the WET record's URL to the keepURLs.txt file or discardURLs.txt file.
|
---|
169 | */
|
---|
170 | private void processWETrecord(BufferedWriter keepURLsWriter, BufferedWriter discardURLsWriter,
|
---|
171 | int recordID, int contentLength, int lineCount,
|
---|
172 | String recordURI, String record)
|
---|
173 | {
|
---|
174 | System.err.println("WET #" + this.WETFileID + " record #" + recordID
|
---|
175 | + " - contentLength: " + contentLength
|
---|
176 | + " - lineCount: " + lineCount);
|
---|
177 | System.err.println("URI: " + recordURI);
|
---|
178 | //System.err.println(record);
|
---|
179 | //System.err.println("--------------------------");
|
---|
180 |
|
---|
181 | String paddedFileName = String.format("%04d.txt", recordID);
|
---|
182 |
|
---|
183 | File parentFolder = null;
|
---|
184 |
|
---|
185 | // want to match "product(s)" but not "production"
|
---|
186 | //if(recordURI.matches(".*/?product[^a-rt-z].*")) {//if(recordURI.matches(".*/?products?/?.*")) {
|
---|
187 |
|
---|
188 |
|
---|
189 | /*
|
---|
190 | if(recordURI.contains("product") && !recordURI.contains("production")) {
|
---|
191 |
|
---|
192 | // don't want a "translated" product site/online store
|
---|
193 | // These curiously often tend to have "product(s)" in the URL
|
---|
194 | parentFolder = batchProcessor.discardFolder;
|
---|
195 | }
|
---|
196 |
|
---|
197 | else if(lineCount >= MIN_LINE_COUNT && contentLength >= MIN_CONTENT_LENGTH) {
|
---|
198 | parentFolder = batchProcessor.keepFolder;
|
---|
199 | System.err.println("@@@KEEPING");
|
---|
200 | } else if(contentLength >= MIN_CONTENT_LENGTH_WRAPPED_LINE) {
|
---|
201 | int countSpaces = 0;
|
---|
202 | for(int i = 0; i < record.length(); i++) {
|
---|
203 | if(record.charAt(i) == ' ') countSpaces++;
|
---|
204 | }
|
---|
205 | if(countSpaces >= MIN_SPACES_IN_A_WRAPPED_LINE) {
|
---|
206 | // So we have at least 500 chars (possibly on a single wrapped line)
|
---|
207 | // containing at least 10 spaces. Such a record is also worth keeping.
|
---|
208 | parentFolder = batchProcessor.keepFolder;
|
---|
209 | }
|
---|
210 | }
|
---|
211 | */
|
---|
212 |
|
---|
213 | if(batchProcessor.isBlacklisted(recordURI)) {
|
---|
214 | parentFolder = batchProcessor.discardFolder;
|
---|
215 | }
|
---|
216 | else if(batchProcessor.isGreylisted(recordURI)) { // e.g. products sites
|
---|
217 | parentFolder = batchProcessor.discardFolder; // TODO: checkfolder
|
---|
218 | } else {
|
---|
219 | // If a web page's WET record contains a certain minimum number of words,
|
---|
220 | // we will think it's a meaningful web page and has sufficient content for text analysis
|
---|
221 | // to have been successful. Cut off values at present are:
|
---|
222 | // - a minimum of 20 words
|
---|
223 | // - a word consists of 1 to 15 chars demarcated by spaces. Any more chars may point
|
---|
224 | // to words having been glued together. This is used by irrelevant sites and moreover
|
---|
225 | // can't be analysed for language, so may not be actually MRI.
|
---|
226 |
|
---|
227 | // Though StringTokenizer still in use, as seen in discussion at
|
---|
228 | // https://stackoverflow.com/questions/6983856/why-is-stringtokenizer-deprecated
|
---|
229 | // String.split(regex) seems better for splitting on general whitespace
|
---|
230 | String[] allWords = record.split("\\s");
|
---|
231 | int validWordCount = 0;
|
---|
232 | int numCamelCaseWords = 0;
|
---|
233 | for(int i = 0; i < allWords.length; i++) {
|
---|
234 | String word = allWords[i];
|
---|
235 |
|
---|
236 | // throw away if n words contain camelcase, which is another case of words glued together
|
---|
237 | if(word.matches(".*[a-z][A-Z].*") && word.length() >= 5) {
|
---|
238 | numCamelCaseWords++;
|
---|
239 | }
|
---|
240 |
|
---|
241 | // In Maori, word length of 1 is not uncommon
|
---|
242 | // but let's skip camelcased words when counting valid words
|
---|
243 | else if(word.length() >= 1 && word.length() <= batchProcessor.MAX_WORD_LENGTH) validWordCount++;
|
---|
244 | }
|
---|
245 |
|
---|
246 | // dump if too many camelcase words (ideally keep none of that kind?)
|
---|
247 | if(numCamelCaseWords >= batchProcessor.MAX_WORDS_CAMELCASE) {
|
---|
248 | parentFolder = batchProcessor.discardFolder;
|
---|
249 | System.err.println("@@@DISCARDING - CAMELCASED CONTENTS");
|
---|
250 | }
|
---|
251 | else if(validWordCount >= batchProcessor.MIN_NUM_WORDS) { // otherwise, keep anything with a sufficient number of valid words
|
---|
252 | parentFolder = batchProcessor.keepFolder;
|
---|
253 | System.err.println("@@@KEEPING");
|
---|
254 | }
|
---|
255 | }
|
---|
256 | // if parentFolder still not set, set to discard pile folder
|
---|
257 | if(parentFolder == null) {
|
---|
258 | parentFolder = batchProcessor.discardFolder;
|
---|
259 | System.err.println("@@@DISCARDING");
|
---|
260 | }
|
---|
261 |
|
---|
262 | try {
|
---|
263 | if (parentFolder == batchProcessor.keepFolder) {
|
---|
264 | keepURLsWriter.write(recordURI + "\n");
|
---|
265 | } else {
|
---|
266 | discardURLsWriter.write(recordURI + "\n");
|
---|
267 | }
|
---|
268 | } catch(Exception e) {
|
---|
269 | System.err.println("Unable to write URL");
|
---|
270 | e.printStackTrace();
|
---|
271 | }
|
---|
272 |
|
---|
273 | System.err.println("--------------------------");
|
---|
274 |
|
---|
275 | File outFile = new File(parentFolder, this.WETFileID + "-" + paddedFileName);
|
---|
276 |
|
---|
277 | try (BufferedWriter writer = new BufferedWriter(new FileWriter(outFile))) {
|
---|
278 | writer.write(record);
|
---|
279 | // Try-with-resources examples don't call close() explicitly:
|
---|
280 | // https://docs.oracle.com/javase/tutorial/essential/exceptions/tryResourceClose.html
|
---|
281 | //writer.close();
|
---|
282 | } catch(IOException ioe) {
|
---|
283 | ioe.printStackTrace();
|
---|
284 | System.err.println("\n@@@@@@@@@ Error writing to file " + outFile);
|
---|
285 | }
|
---|
286 | }
|
---|
287 | }
|
---|