first commit
This commit is contained in:
Peter Kleiweg
9
NieuwsNL/Makefile
Normal file
9
NieuwsNL/Makefile
Normal file
@@ -0,0 +1,9 @@
|
||||
all: \
|
||||
metadata \
|
||||
nieuwsnl
|
||||
|
||||
metadata: cmd/metadata/*.go
|
||||
go build -o $@ $^
|
||||
|
||||
nieuwsnl: cmd/nieuwsnl/*.go
|
||||
go build -o $@ $^
|
||||
126
NieuwsNL/cmd/metadata/metadata.go
Normal file
126
NieuwsNL/cmd/metadata/metadata.go
Normal file
@@ -0,0 +1,126 @@
|
||||
package main
|
||||
|
||||
import (
|
||||
"github.com/pebbe/util"
|
||||
|
||||
"bufio"
|
||||
"encoding/xml"
|
||||
"fmt"
|
||||
"html"
|
||||
"os"
|
||||
"strings"
|
||||
"time"
|
||||
)
|
||||
|
||||
type Item struct {
|
||||
XMLName xml.Name `xml:"item"`
|
||||
UnixTime int64 `xml:"unixTime"`
|
||||
}
|
||||
|
||||
var (
|
||||
x = util.CheckErr
|
||||
escape = html.EscapeString
|
||||
data = make(map[string][]string)
|
||||
location *time.Location
|
||||
)
|
||||
|
||||
func main() {
|
||||
var err error
|
||||
location, err = time.LoadLocation("Europe/Amsterdam")
|
||||
x(err)
|
||||
|
||||
files, err := os.ReadDir("..")
|
||||
x(err)
|
||||
for _, file := range files {
|
||||
filename := file.Name()
|
||||
if strings.HasSuffix(filename, ".txt") {
|
||||
doText(filename)
|
||||
} else if strings.HasSuffix(filename, ".xml") {
|
||||
doXml(filename)
|
||||
}
|
||||
}
|
||||
|
||||
files, err = os.ReadDir("xml")
|
||||
x(err)
|
||||
for _, file := range files {
|
||||
filename := file.Name()
|
||||
if !strings.HasSuffix(filename, ".xml") {
|
||||
continue
|
||||
}
|
||||
aa := strings.Split(filename, ".")
|
||||
base := strings.Join(aa[1:len(aa)-2], ".")
|
||||
b, err := os.ReadFile("xml/" + filename)
|
||||
x(err)
|
||||
s := string(b)
|
||||
i := strings.Index(s, "<alpino") + 1
|
||||
i += strings.Index(s[i:], "<")
|
||||
fp, err := os.Create("xml/" + filename + ".tmp")
|
||||
x(err)
|
||||
_, err = fp.WriteString(s[:i])
|
||||
x(err)
|
||||
_, err = fp.WriteString("<metadata>\n <meta type=\"text\" name=\"source\" value=\"NieuwsNL\"/>\n")
|
||||
x(err)
|
||||
for _, m := range data[base] {
|
||||
_, err = fp.WriteString(" " + m + "\n")
|
||||
x(err)
|
||||
}
|
||||
_, err = fp.WriteString(" </metadata>\n ")
|
||||
x(err)
|
||||
_, err = fp.WriteString(stripMeta(s[i:]))
|
||||
x(err)
|
||||
x(fp.Close())
|
||||
x(os.Rename("xml/"+filename+".tmp", "xml/"+filename))
|
||||
}
|
||||
}
|
||||
|
||||
func doText(filename string) {
|
||||
base := filename[:len(filename)-4]
|
||||
if _, ok := data[base]; !ok {
|
||||
data[base] = make([]string, 0)
|
||||
}
|
||||
fp, err := os.Open("../" + filename)
|
||||
x(err)
|
||||
defer func() { x(fp.Close()) }()
|
||||
scanner := bufio.NewScanner(fp)
|
||||
for scanner.Scan() {
|
||||
line := scanner.Text()
|
||||
if !strings.HasPrefix(line, "##META") {
|
||||
continue
|
||||
}
|
||||
aa := strings.Fields(line)
|
||||
if len(aa) > 4 {
|
||||
data[base] = append(data[base],
|
||||
fmt.Sprintf(`<meta type="%s" name="%s" value="%s"/>`,
|
||||
aa[1],
|
||||
escape(aa[2]),
|
||||
escape(strings.Join(aa[4:], " "))))
|
||||
}
|
||||
}
|
||||
x(scanner.Err())
|
||||
}
|
||||
|
||||
func doXml(filename string) {
|
||||
base := filename[:len(filename)-4]
|
||||
if _, ok := data[base]; !ok {
|
||||
data[base] = make([]string, 0)
|
||||
}
|
||||
b, err := os.ReadFile("../" + filename)
|
||||
x(err)
|
||||
var item Item
|
||||
x(xml.Unmarshal(b, &item))
|
||||
t := time.Unix(item.UnixTime, 0).In(location)
|
||||
data[base] = append(data[base],
|
||||
fmt.Sprintf(`<meta type="date" name="pubdate" value="%d-%02d-%02d"/>`,
|
||||
t.Year(),
|
||||
int(t.Month()),
|
||||
t.Day()))
|
||||
}
|
||||
|
||||
func stripMeta(s string) string {
|
||||
i1 := strings.Index(s, "<metadata>")
|
||||
if i1 < 0 {
|
||||
return s
|
||||
}
|
||||
i2 := i1 + strings.Index(s[i1:], "</metadata>") + 11
|
||||
return s[:i1] + strings.TrimLeft(s[i2:], " \t\r\n")
|
||||
}
|
||||
217
NieuwsNL/cmd/nieuwsnl/nieuwsnl.go
Normal file
217
NieuwsNL/cmd/nieuwsnl/nieuwsnl.go
Normal file
@@ -0,0 +1,217 @@
|
||||
package main
|
||||
|
||||
import (
|
||||
"github.com/jbowtie/gokogiri"
|
||||
"github.com/pebbe/util"
|
||||
|
||||
"bytes"
|
||||
"encoding/xml"
|
||||
"fmt"
|
||||
"io"
|
||||
"net/http"
|
||||
"net/url"
|
||||
"os"
|
||||
"strings"
|
||||
"time"
|
||||
)
|
||||
|
||||
type Rss struct {
|
||||
XMLName xml.Name `xml:"rss"`
|
||||
Items []ItemT `xml:"channel>item"`
|
||||
}
|
||||
|
||||
type ItemT struct {
|
||||
Title string `xml:"title"`
|
||||
PubDate string `xml:"pubDate"`
|
||||
UnixTime int64 `xml:"unixTime"`
|
||||
Guid string `xml:"guid"`
|
||||
Link string `xml:"link"`
|
||||
Data []byte `xml:",innerxml"`
|
||||
}
|
||||
|
||||
var (
|
||||
x = util.CheckErr
|
||||
w = util.WarnErr
|
||||
agent = "AhrefsBot/7.0"
|
||||
)
|
||||
|
||||
func exists(filename string) bool {
|
||||
_, err := os.Stat(filename)
|
||||
return err == nil
|
||||
}
|
||||
|
||||
func fileDate(filename string) string {
|
||||
b, err := os.ReadFile(filename)
|
||||
if err != nil {
|
||||
return ""
|
||||
}
|
||||
s := string(b)
|
||||
i1 := strings.Index(s, "<unixTime>") + 10
|
||||
i2 := strings.Index(s, "</unixTime>")
|
||||
return s[i1:i2]
|
||||
}
|
||||
|
||||
func main() {
|
||||
req, err := http.NewRequest("GET", "https://nieuws.nl/sitemap/news.xml", nil)
|
||||
x(err)
|
||||
req.Header.Set("User-Agent", agent)
|
||||
|
||||
client := &http.Client{}
|
||||
resp, err := client.Do(req)
|
||||
x(err)
|
||||
body, err := io.ReadAll(resp.Body)
|
||||
x(err)
|
||||
x(resp.Body.Close())
|
||||
|
||||
var rss Rss
|
||||
x(xml.Unmarshal(body, &rss))
|
||||
|
||||
if len(rss.Items) == 0 {
|
||||
x(fmt.Errorf("len(rss.Items) == 0"))
|
||||
}
|
||||
|
||||
for _, item := range rss.Items {
|
||||
t, err := time.Parse(time.RFC1123Z, item.PubDate)
|
||||
if err != nil {
|
||||
t, err = time.Parse(time.RFC1123, item.PubDate)
|
||||
}
|
||||
x(err)
|
||||
dirname := fmt.Sprintf("/net/corpora/nlnieuws/NieuwsNL/%d/%02d/%02d", t.Year(), int(t.Month()), t.Day())
|
||||
filename := dirname + "/" + url.PathEscape(strings.TrimPrefix(item.Guid, "urn:uuid:"))
|
||||
|
||||
ts := fmt.Sprintf("%d", t.Unix())
|
||||
needUpdate := fileDate(filename+".xml") != ts
|
||||
|
||||
x(os.MkdirAll(dirname, 0777))
|
||||
fp, err := os.Create(filename + ".xml")
|
||||
x(err)
|
||||
_, err = fp.WriteString("<?xml version=\"1.0\" encoding=\"UTF-8\"?>\n<item>\n")
|
||||
x(err)
|
||||
_, err = fmt.Fprintf(fp, "<unixTime>%d</unixTime>", t.Unix())
|
||||
x(err)
|
||||
_, err = fp.Write(item.Data)
|
||||
x(err)
|
||||
_, err = fp.WriteString("</item>\n")
|
||||
x(err)
|
||||
x(fp.Close())
|
||||
x(os.Chtimes(filename+".xml", t, t))
|
||||
if !doArticle(filename, item.Link, item.Title, t, needUpdate) {
|
||||
x(os.Remove(filename + ".xml"))
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
func doArticle(filename string, url string, title string, timestamp time.Time, needUpdate bool) bool {
|
||||
if needUpdate {
|
||||
_ = os.Remove(filename + ".err")
|
||||
_ = os.Remove(filename + ".html")
|
||||
_ = os.Remove(filename + ".txt")
|
||||
_ = os.Remove(filename + ".skip")
|
||||
} else {
|
||||
if exists(filename+".txt") || exists(filename+".skip") {
|
||||
return true
|
||||
}
|
||||
}
|
||||
time.Sleep(2 * time.Second)
|
||||
|
||||
req, err := http.NewRequest("GET", url, nil)
|
||||
x(err)
|
||||
req.Header.Set("User-Agent", agent)
|
||||
|
||||
client := &http.Client{}
|
||||
resp, err := client.Do(req)
|
||||
x(err)
|
||||
body, err := io.ReadAll(resp.Body)
|
||||
x(err)
|
||||
x(resp.Body.Close())
|
||||
|
||||
doc, err := gokogiri.ParseHtml(body)
|
||||
x(err)
|
||||
|
||||
var buf bytes.Buffer
|
||||
fouten := make([]string, 0)
|
||||
|
||||
root := doc.Root()
|
||||
|
||||
var cat string
|
||||
aa, err := root.Search(`//a[contains(@class, "articleHeader__info__category")]`)
|
||||
x(err)
|
||||
if len(aa) == 0 {
|
||||
_, err = fmt.Fprintln(&buf, "##META text cat =")
|
||||
x(err)
|
||||
_ = w(fmt.Errorf("no cat: %s", url))
|
||||
// geen fout, maar waarschuwing als er meer fouten zijn
|
||||
fouten = append(fouten, fmt.Sprintf("no text: %s\n", url))
|
||||
// dus geen return false
|
||||
} else {
|
||||
for _, a := range aa {
|
||||
cat = strings.ReplaceAll(a.Content(), "\n", " ")
|
||||
_, err = fmt.Fprintf(&buf, "##META text cat = %s\n", cat)
|
||||
x(err)
|
||||
}
|
||||
}
|
||||
|
||||
_, err = buf.WriteString(addEnd(title))
|
||||
x(err)
|
||||
|
||||
// oud: //div[@id="article-blocks"]//p
|
||||
pp, err := root.Search(`//div[@id="article-blocks"]//div[contains(@class, "paragraph-content")]`)
|
||||
x(err)
|
||||
if len(pp) == 0 {
|
||||
_ = w(fmt.Errorf("empty: %s", url))
|
||||
// dit is echt fout
|
||||
fouten = append(fouten, fmt.Sprintf("empty: %s\n", url))
|
||||
|
||||
fp, err := os.Create(filename + ".err")
|
||||
x(err)
|
||||
for _, fout := range fouten {
|
||||
_, err = fp.WriteString(fout)
|
||||
x(err)
|
||||
}
|
||||
x(fp.Close())
|
||||
x(os.Chtimes(filename+".err", timestamp, timestamp))
|
||||
|
||||
fp, err = os.Create(filename + ".html")
|
||||
x(err)
|
||||
_, err = fp.Write(body)
|
||||
x(err)
|
||||
x(fp.Close())
|
||||
x(os.Chtimes(filename+".html", timestamp, timestamp))
|
||||
|
||||
return false // echt fout
|
||||
}
|
||||
for _, p := range pp {
|
||||
_, err = buf.WriteString(addEnd(p.Content()))
|
||||
x(err)
|
||||
}
|
||||
|
||||
fp, err := os.Create(filename + ".txt")
|
||||
x(err)
|
||||
_, err = fp.Write(buf.Bytes())
|
||||
x(err)
|
||||
x(fp.Close())
|
||||
|
||||
x(os.Chtimes(filename+".txt", timestamp, timestamp))
|
||||
|
||||
return true
|
||||
}
|
||||
|
||||
func addEnd(s string) string {
|
||||
s = strings.TrimSpace(s)
|
||||
n := len(s)
|
||||
if n == 0 {
|
||||
return ""
|
||||
}
|
||||
if n > 0 {
|
||||
if strings.ContainsAny(s[n-1:], ".!?") {
|
||||
return s + "\n"
|
||||
}
|
||||
}
|
||||
if n > 1 {
|
||||
s2 := s[n-2:]
|
||||
if s2 == `."` || s2 == `!"` || s2 == `?"` || s2 == `.'` || s2 == `!'` || s2 == `?'` {
|
||||
return s + "\n"
|
||||
}
|
||||
}
|
||||
return s + ".\n"
|
||||
}
|
||||
65
NieuwsNL/txt2corpus.sh
Executable file
65
NieuwsNL/txt2corpus.sh
Executable file
@@ -0,0 +1,65 @@
|
||||
#!/bin/bash
|
||||
|
||||
set -e
|
||||
|
||||
unset CDPATH
|
||||
PATH=/net/corpora/nlnieuws/bin:/net/aps/bin:$PATH
|
||||
export TZ=Europe/Amsterdam
|
||||
. /net/aps/etc/alpino-activate.sh > /dev/null
|
||||
|
||||
if [ "$1" = "" ]
|
||||
then
|
||||
# nieuws.nl gaat per dag, niet per week
|
||||
# dus gegevens van 2 dagen geleden, niet een week geleden
|
||||
ds=`ISODate -2`
|
||||
else
|
||||
case "$1" in
|
||||
2[0-9][0-9][0-9]-[01][0-9]-[0-3][0-9])
|
||||
ds=$1
|
||||
;;
|
||||
*)
|
||||
echo INVALID
|
||||
exit 1
|
||||
;;
|
||||
esac
|
||||
fi
|
||||
|
||||
dp=${ds//-//}
|
||||
|
||||
corpus=/net/corpora/nlnieuws/NieuwsNL/corpus/$ds
|
||||
|
||||
cd /net/corpora/nlnieuws/NieuwsNL/$dp
|
||||
|
||||
ln -s lock.$$ lock
|
||||
if [ "`readlink lock`" != lock.$$ ]
|
||||
then
|
||||
echo Getting lock failed
|
||||
exit 1
|
||||
fi
|
||||
|
||||
rm -fr out
|
||||
mkdir out
|
||||
|
||||
rm -f $corpus.lines
|
||||
for i in *.txt
|
||||
do
|
||||
b=`basename $i .txt`
|
||||
perl -p -e 's/^\s*//; s/^##META.*\n//' $i | tokenize.sh \
|
||||
| perl -e '$n = 0; while(<>) { $n++; print("nnl.'$b'.$n|$_"); }' \
|
||||
>> $corpus.lines
|
||||
done
|
||||
|
||||
cd out
|
||||
mkdir xml
|
||||
Alpino -flag treebank xml debug=1 end_hook=xml user_max=900000 -parse < $corpus.lines 2> $corpus.log
|
||||
|
||||
../../../../metadata 2> err
|
||||
rm err
|
||||
|
||||
cd xml
|
||||
alto -o $corpus.data.dz *.xml 2> /dev/null
|
||||
|
||||
cd ../..
|
||||
rm -fr out
|
||||
|
||||
rm -f lock
|
||||
Reference in New Issue
Block a user