File size: 38,785 Bytes
6fa4bc9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
{
    "paper_id": "2021",
    "header": {
        "generated_with": "S2ORC 1.0.0",
        "date_generated": "2023-01-19T03:35:49.982304Z"
    },
    "title": "Interesting cross-border news discovery using cross-lingual article linking and document similarity",
    "authors": [
        {
            "first": "Boshko",
            "middle": [],
            "last": "Koloski",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "Jo\u017eef Stefan Institute Jo\u017eef Stefan IPS",
                "location": {
                    "addrLine": "Jamova 39",
                    "settlement": "Ljubljana"
                }
            },
            "email": "boshko.koloski@ijs.si"
        },
        {
            "first": "Elaine",
            "middle": [],
            "last": "Zosa",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "University of Helsinki",
                "location": {
                    "addrLine": "Pietari Kalmin katu 5",
                    "settlement": "Helsinki"
                }
            },
            "email": "elaine.zosa@helsinki.fi"
        },
        {
            "first": "Timen",
            "middle": [],
            "last": "Stepi\u0161nik-Perdih",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "Jo\u017eef Stefan Institute",
                "location": {
                    "addrLine": "Jamova 39, Ljubljana Bla\u017e\u0160krlj Jo\u017eef Stefan Institute Jamova 39",
                    "settlement": "Ljubljana"
                }
            },
            "email": ""
        },
        {
            "first": "Tarmo",
            "middle": [],
            "last": "Paju",
            "suffix": "",
            "affiliation": {},
            "email": "tarmo.paju@delfi.ee"
        },
        {
            "first": "Senja",
            "middle": [],
            "last": "Pollak",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "Jo\u017eef Stefan Institute",
                "location": {
                    "addrLine": "Jamova 39",
                    "settlement": "Ljubljana"
                }
            },
            "email": "senja.pollak@ijs.si"
        }
    ],
    "year": "",
    "venue": null,
    "identifiers": {},
    "abstract": "Contemporary news media face increasing amounts of available data that can be of use when prioritizing, selecting and discovering new news. In this work we propose a methodology for retrieving interesting articles in a cross-border news discovery setting. More specifically, we explore how a set of seed documents in Estonian can be projected in Latvian document space and serve as a basis for discovery of novel interesting pieces of Latvian news that would interest Estonian readers. The proposed methodology was evaluated by Estonian journalist who confirmed that in the best setting, from top 10 retrieved Latvian documents, half of them represent news that are potentially interesting to be taken by the Estonian media house and presented to Estonian readers.",
    "pdf_parse": {
        "paper_id": "2021",
        "_pdf_hash": "",
        "abstract": [
            {
                "text": "Contemporary news media face increasing amounts of available data that can be of use when prioritizing, selecting and discovering new news. In this work we propose a methodology for retrieving interesting articles in a cross-border news discovery setting. More specifically, we explore how a set of seed documents in Estonian can be projected in Latvian document space and serve as a basis for discovery of novel interesting pieces of Latvian news that would interest Estonian readers. The proposed methodology was evaluated by Estonian journalist who confirmed that in the best setting, from top 10 retrieved Latvian documents, half of them represent news that are potentially interesting to be taken by the Estonian media house and presented to Estonian readers.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Abstract",
                "sec_num": null
            }
        ],
        "body_text": [
            {
                "text": "This paper presents our results of the participation in the hackaton, which was organised as part of the EACL 2021 Hackashop on news media content analysis and automated report generation. We are addressing the EMBEDDIA hackathon challenge on Identifying Interesting News from Neighbouring Countries (Pollak et al., 2021) in Estonian and Latvian context, which is a fully novel document retrieval task performed on recently released EM-BEDDIA news datasets. Estonian journalists are very interested in identifying stories from Latvia, which will attract a large number of readers and are \"special\". While performing keyword-based search for Latvian news, where Estonians are mentioned is a simple task, this challenge on the contrary aims to identify a small set of documents from a larger number of topics, e.g. scandals, deaths and gossip that might be somehow connected to Estonia: not only by mentioning Estonians but by identifying news and stories that Estonians relate to (for example, when similar things have happened in Estonia or when similar news have been popular in Estonia).",
                "cite_spans": [
                    {
                        "start": 300,
                        "end": 321,
                        "text": "(Pollak et al., 2021)",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "In our approach, we first automatically create a collection of interesting articles using a stringbased search and cross-lingual document linking, and then rank the query documents based on the proportion of interesting documents in their neighbourhood (where the neighbourhood is defined by a document similarity) by the newly introduced Seed news of interest score (SNIR).",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "The article first presents the datasets (Section 2), introduces the methodology (Section 3), and presents our experimental results (Section 4). The code and the data are made publicly available (see Section 5). Finally, Section 6 concludes the paper and presents the ideas for further work.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "In this study, we used the following resources.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Datasets",
                "sec_num": "2"
            },
            {
                "text": "\u2022 Archives of Estonian news articles from Ekspress Meedia. Ekspress Meedia belongs to Ekspress Meedia Group, one of the largest media groups in the Baltics. From the entire collection of Ekspress Meedia articles (Pollak et al., 2021) , we selected the articles from the years 2018 and 2019 (i.e. 64,651 articles in total).",
                "cite_spans": [
                    {
                        "start": 212,
                        "end": 233,
                        "text": "(Pollak et al., 2021)",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Datasets",
                "sec_num": "2"
            },
            {
                "text": "\u2022 Dataset of archives of Latvian news articles (Pollak et al., 2021) come from Latvian Delfi that also belongs to Ekspress Meedia Group. We considered only the articles from the years 2018 and 2019 (i.e. 60,802 articles in total).",
                "cite_spans": [
                    {
                        "start": 47,
                        "end": 68,
                        "text": "(Pollak et al., 2021)",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Datasets",
                "sec_num": "2"
            },
            {
                "text": "\u2022 Manually identified interesting news for Estonian readers in Latvian (and their Estonian counterparts). These were manually identified as examples of interesting news by Estonian journalist from Ekspress Meedia. Note that the Estonian articles are not their direct translations, as the articles can be slightly adapted to Estonian audience.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Datasets",
                "sec_num": "2"
            },
            {
                "text": "Our methodology consists of two steps. First, we automatically construct the datasets of interesting Latvian articles and next propose a method to retrieve interesting articles by ranking a given query document based on the the proportion of interesting articles in its neighbourhood.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Methodology",
                "sec_num": "3"
            },
            {
                "text": "The aim of this step is to automatically construct Latvian seed news of interest, which are considered as good examples of interesting Latvian articles.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Automated selection of Latvian example articles",
                "sec_num": "3.1"
            },
            {
                "text": "As there are only 21 manually identified examples, which we keep for the evaluation purposes and parameter setting, this step was automatised.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Automated selection of Latvian example articles",
                "sec_num": "3.1"
            },
            {
                "text": "In our approach, we first extract Estonian articles, that specifically mention the source of Latvian Delfi (L\u00e4ti Delfi), which leads to 100 identified Estonian articles which are considered as automatically constructed Estonian example data. Then we follow the methodology of Zosa et al. (2020) . More specifically, we use Sentence-BERT (Reimers and Gurevych, 2019) to obtain cross-lingual encodings of the articles from both languages. For each Estonian article, we extract best k Latvian candidates (where k is a parameter) by taking the cosine similarities between the query Estonian article and all the candidate Latvian articles and finally rank the Latvian articles based on this similarity measure.",
                "cite_spans": [
                    {
                        "start": 276,
                        "end": 294,
                        "text": "Zosa et al. (2020)",
                        "ref_id": "BIBREF4"
                    },
                    {
                        "start": 337,
                        "end": 365,
                        "text": "(Reimers and Gurevych, 2019)",
                        "ref_id": "BIBREF3"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Automated selection of Latvian example articles",
                "sec_num": "3.1"
            },
            {
                "text": "Note that also recent work (Litschko et al., 2021) has shown that specialized sentence encoders trained for semantic similarity across languages obtain better results in document retrieval than static cross-lingual word embeddings or averaged contextualized embeddings.",
                "cite_spans": [
                    {
                        "start": 27,
                        "end": 50,
                        "text": "(Litschko et al., 2021)",
                        "ref_id": "BIBREF0"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Automated selection of Latvian example articles",
                "sec_num": "3.1"
            },
            {
                "text": "In this step, we assign the \"interestingness score\" to each query article. First, we identify the local neighbourhood of a query article by document similarity. We use the same sentence-embeddings method as in the previous step, with the difference that here the articles similarity is computed in monolingual setting. The number of articles surrounding the query is a parameter m.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Retrieval of interesting news articles",
                "sec_num": "3.2"
            },
            {
                "text": "We introduce a custom metric called SNIR (Seed news of interest ratio), where we compute the ratio of automatically identified interesting news compared to all the articles in the neighbourhood. The hypothesis is that the articles of interest will have more articles from the automatically identified interesting news articles in their surrounding than the articles, which are not relevant for the Estonian readers.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Retrieval of interesting news articles",
                "sec_num": "3.2"
            },
            {
                "text": "The result of our method is a ranked list of articles for a given time period (e.g. a day, week, month) where a journalist can then decide to manually check top x articles. In addition, in future also a SNIR threshold could be set which would allow interested journalists to be informed about potentially interesting articles in real-time.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Retrieval of interesting news articles",
                "sec_num": "3.2"
            },
            {
                "text": "The SNIR score is defined as follows. Let NeighborhoodDocuments m represent the set of m nearest documents in the final embedding space. Let Interesting m represent the set of m interesting seed documents obtained via the cross-lingual mapping discussed in the previous sections. We can define the SNIR at m as:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Retrieval of interesting news articles",
                "sec_num": "3.2"
            },
            {
                "text": "SNIR(m) = |Interesting m | |NeighborhoodDocuments m | .",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Retrieval of interesting news articles",
                "sec_num": "3.2"
            },
            {
                "text": "We report SNIR values for different neighborhood (m) sizes. The goal of SNIR is to score interesting query articles higher than query articles which are not of special interest for Estonian readers.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Retrieval of interesting news articles",
                "sec_num": "3.2"
            },
            {
                "text": "4 Experiments and results",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Retrieval of interesting news articles",
                "sec_num": "3.2"
            },
            {
                "text": "For our experiments, we used the following settings. We test the parameter setting for k in cross-lingual article linking to 20 and 100, and the setting of parameter m to 10, 20 and 100 for determining the neighbourhood in computing the SNIR score. First, we evaluated the cross-lingual article linking on the 21 manually linked article pairs. For these article pairs, We obtained an MRR (Mean Reciprocal Rank) score of 64.93%, which shows that for an article in a source language, the correct article is usually proposed as the first or second candidate.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Automated analysis",
                "sec_num": "4.1"
            },
            {
                "text": "Next, we performed qualitative analysis by visualising the document space. In Figure 1 (using parameter k=20), we can see that automatically defined Latvian seed news of interest (red) are not evenly distributed and support the hypothesis, that random article's neighbourhood will differ in this respect. The figure also presents the manually identified interesting news (orange), where at least some of the documents seem to be positioned together.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 78,
                        "end": 86,
                        "text": "Figure 1",
                        "ref_id": "FIGREF1"
                    }
                ],
                "eq_spans": [],
                "section": "Automated analysis",
                "sec_num": "4.1"
            },
            {
                "text": "Next, we compare the SNIR scores of 21 manually identified interesting articles compared to random Latvian articles. The results of SNIR score for parameter k=100 at different m can be found in Figure 2 . This also suggests that there is some evidence that a threshold could in future be determined, but more extensive experiments should be performed in future work.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 194,
                        "end": 202,
                        "text": "Figure 2",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Automated analysis",
                "sec_num": "4.1"
            },
            {
                "text": "For final evaluation, we selected the last month of the Latvian collection (1408 articles in total), and ranked the articles according to the SNIR score. These were provided to the Estonian journalist of Ekspress Meedia who evaluated top 10 results for each of the settings.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Manual analysis",
                "sec_num": "4.2"
            },
            {
                "text": "We prepared four different pairs of k \u2208 10, 100 retrieved documents and m \u2208 10, 20, 100 documents in the neighbourhood which were evaluated by the media expert. The media house expert analyzed the retrieved documents and labeled them with three different labels based on the acceptance:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Manual analysis",
                "sec_num": "4.2"
            },
            {
                "text": "\u2022 No -the article was of not relevant significance to the media house.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Manual analysis",
                "sec_num": "4.2"
            },
            {
                "text": "\u2022 Maybe -the article contained news about events that are potentially relevant to the Estonian readers.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Manual analysis",
                "sec_num": "4.2"
            },
            {
                "text": "\u2022 Yes -the article contained news about events that are relevant to the Estonian readers or contained extraordinarily news.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Manual analysis",
                "sec_num": "4.2"
            },
            {
                "text": "The evaluation of the top 10 articles retrieved for each k, m pair is listed in Table 1 . From the evaluation we can see that when we have a relatively small number of retrieved documents and a smaller neighbourhood we can benefit from the SNIR metric. As the best performing parameter pairs were the k = 20 and m = 10 retrieving 50% articles as relevant or of close relevance to the Estonian news house. When larger neighbourhood is introduced the space becomes sparser and the method tends to retrieve more false positives. k m Yes Maybe Not  20  10  2  3  5  20 100  1  3  6  100 20  1  3  6  100 100  0  3  7   Table 1 : Evaluation of the top-10 retrieved articles by the SNIR ranking for various k interesting Latvian seeds documents and m neighbourhood sizes.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 80,
                        "end": 87,
                        "text": "Table 1",
                        "ref_id": null
                    },
                    {
                        "start": 534,
                        "end": 622,
                        "text": "Maybe Not  20  10  2  3  5  20 100  1  3  6  100 20  1  3  6  100 100  0  3  7   Table 1",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Manual analysis",
                "sec_num": "4.2"
            },
            {
                "text": "The journalist also explained why a selected news example from positive category is very relevant. The news talks about a scooter accident in court proceedings, which is extraordinary, as well as relevant to Estonians as the debate around scooters at the streets is also very active in Estonia. Some examples from negative category contain articles about foreign news (terror attack, for example) and these are not the type of news that the Estonian journalists would pick from Latvian media.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Manual analysis",
                "sec_num": "4.2"
            },
            {
                "text": "The code and data of the experiments is made available on the GitHub: https://github.com/bkolo sk1/Interesting-cross-border-news-discov ery 6 Conclusion and future work",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Availability",
                "sec_num": "5"
            },
            {
                "text": "In this work we tackled the problem of retrieving interesting news from one country for the context of another neighbouring country. We focused on finding interesting news in Latvian news space that would be engaging for the Estonian public. We used Latvian and Estonian EMBEDDIA datasets to construct the document space. First we used a string matching approach to identify a subset of news in Estonian media that originated from Latvian news. Next, we utilized the methods for ad hoc Cross Lingual document retrieval to find corresponding articles in the Latvian news space. After automatically retrieving this set of Latvian news articles of interest, we used this information in a novel metric defined as SNIR, that analyses a news article's neighbourhood in order to measure it's relevance (interestingness). The assumption of the metric is that if the surrounding documents of a query point are relevant, this new point might be of relevance. The SNIR scores of randomly selected 20 documents and 20 documents identified as examples of interesting news by an Estonian journalist showed that their value differ, which is Figure 2 : Evaluation of the SNIR metric for the 21 gold standard queries (manually identified news of interest) and 21 random query points. The results indicate that a random documents' neighbourhood is structured differently compared to the one of relevant interesting documents.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 1126,
                        "end": 1134,
                        "text": "Figure 2",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Availability",
                "sec_num": "5"
            },
            {
                "text": "promising. Finally, we prepared a test set of news from one month and sent them to manual evaluation by a journalist. Results of top 10 candidates of each setting suggest that the proposed metric works well if the parameters of interesting articles and neighborhood were adjusted right, with the best performing parameter tuple yielding 50% hit-ratio.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Availability",
                "sec_num": "5"
            },
            {
                "text": "For the further work we propose exploring the keywords appearing in the clusters of interesting news and exploiting their named entity tags in order to achieve even better performance. We also want to include background knowledge from knowledge graphs to improve the document similarity evaluation. Special attention will also be paid to setting a threshold for SNIR which would allow for real-time investigation of best candidates in a real journalistic practice.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Availability",
                "sec_num": "5"
            }
        ],
        "back_matter": [
            {
                "text": "The work was supported by European Union's Horizon 2020 research and innovation programme under grant agreement No 825153, project EM-BEDDIA (Cross-Lingual Embeddings for Less-Represented Languages in European News Media). The work was also supported by the Slovenian Research Agency (ARRS) through core research programme Knowledge Technologies (P2-0103) and the work by B.\u0160. through the ARRS young researcher grant.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Acknowledgements",
                "sec_num": "7"
            }
        ],
        "bib_entries": {
            "BIBREF0": {
                "ref_id": "b0",
                "title": "Evaluating multilingual text encoders for unsupervised cross-lingual retrieval",
                "authors": [
                    {
                        "first": "Robert",
                        "middle": [],
                        "last": "Litschko",
                        "suffix": ""
                    },
                    {
                        "first": "Ivan",
                        "middle": [],
                        "last": "Vuli\u0107",
                        "suffix": ""
                    },
                    {
                        "first": "Simone",
                        "middle": [
                            "Paolo"
                        ],
                        "last": "Ponzetto",
                        "suffix": ""
                    },
                    {
                        "first": "Goran",
                        "middle": [],
                        "last": "Glava\u0161",
                        "suffix": ""
                    }
                ],
                "year": 2021,
                "venue": "Proceedings of ECIR",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Robert Litschko, Ivan Vuli\u0107, Simone Paolo Ponzetto, and Goran Glava\u0161. 2021. Evaluating multilin- gual text encoders for unsupervised cross-lingual re- trieval. In Proceedings of ECIR.",
                "links": null
            },
            "BIBREF2": {
                "ref_id": "b2",
                "title": "EMBEDDIA tools, datasets and challenges: Resources and hackathon contributions",
                "authors": [
                    {
                        "first": "Ivar",
                        "middle": [],
                        "last": "Salmela",
                        "suffix": ""
                    },
                    {
                        "first": "Tarmo",
                        "middle": [],
                        "last": "Krustok",
                        "suffix": ""
                    },
                    {
                        "first": "Carl-Gustav",
                        "middle": [],
                        "last": "Paju",
                        "suffix": ""
                    },
                    {
                        "first": "Leo",
                        "middle": [],
                        "last": "Linden",
                        "suffix": ""
                    },
                    {
                        "first": "Elaine",
                        "middle": [],
                        "last": "Lepp\u00e4nen",
                        "suffix": ""
                    },
                    {
                        "first": "Matej",
                        "middle": [],
                        "last": "Zosa",
                        "suffix": ""
                    },
                    {
                        "first": "Linda",
                        "middle": [],
                        "last": "Ul\u010dar",
                        "suffix": ""
                    },
                    {
                        "first": "Silver",
                        "middle": [],
                        "last": "Freienthal",
                        "suffix": ""
                    },
                    {
                        "first": "Luis",
                        "middle": [
                            "Adri\u00e1n"
                        ],
                        "last": "Traat",
                        "suffix": ""
                    },
                    {
                        "first": "Matej",
                        "middle": [],
                        "last": "Cabrera-Diego",
                        "suffix": ""
                    },
                    {
                        "first": "Nada",
                        "middle": [],
                        "last": "Martinc",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Lavra\u010d",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Bla\u017e\u0161krlj",
                        "suffix": ""
                    },
                    {
                        "first": "Andra\u017e",
                        "middle": [],
                        "last": "Mar-Tin\u017enidar\u0161i\u010d",
                        "suffix": ""
                    },
                    {
                        "first": "Boshko",
                        "middle": [],
                        "last": "Pelicon",
                        "suffix": ""
                    },
                    {
                        "first": "Vid",
                        "middle": [],
                        "last": "Koloski",
                        "suffix": ""
                    },
                    {
                        "first": "Janez",
                        "middle": [],
                        "last": "Podpe\u010dan",
                        "suffix": ""
                    },
                    {
                        "first": "Shane",
                        "middle": [],
                        "last": "Kranjc",
                        "suffix": ""
                    },
                    {
                        "first": "Emanuela",
                        "middle": [],
                        "last": "Sheehan",
                        "suffix": ""
                    },
                    {
                        "first": "Jose",
                        "middle": [],
                        "last": "Boros",
                        "suffix": ""
                    },
                    {
                        "first": "Antoine",
                        "middle": [],
                        "last": "Moreno",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Doucet",
                        "suffix": ""
                    }
                ],
                "year": null,
                "venue": "Proceedings of the EACL Hackashop on News Media Content Analysis and Automated Report Generation",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Salmela, Ivar Krustok, Tarmo Paju, Carl-Gustav Linden, Leo Lepp\u00e4nen, Elaine Zosa, Matej Ul\u010dar, Linda Freienthal, Silver Traat, Luis Adri\u00e1n Cabrera- Diego, Matej Martinc, Nada Lavra\u010d, Bla\u017e\u0160krlj, Mar- tin\u017dnidar\u0161i\u010d, Andra\u017e Pelicon, Boshko Koloski, Vid Podpe\u010dan, Janez Kranjc, Shane Sheehan, Emanuela Boros, Jose Moreno, Antoine Doucet, and Hannu Toivonen. 2021. EMBEDDIA tools, datasets and challenges: Resources and hackathon contributions. In Proceedings of the EACL Hackashop on News Me- dia Content Analysis and Automated Report Gener- ation. Association for Computational Linguistics.",
                "links": null
            },
            "BIBREF3": {
                "ref_id": "b3",
                "title": "Sentence-BERT: Sentence embeddings using Siamese BERTnetworks",
                "authors": [
                    {
                        "first": "Nils",
                        "middle": [],
                        "last": "Reimers",
                        "suffix": ""
                    },
                    {
                        "first": "Iryna",
                        "middle": [],
                        "last": "Gurevych",
                        "suffix": ""
                    }
                ],
                "year": 2019,
                "venue": "Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP)",
                "volume": "",
                "issue": "",
                "pages": "3982--3992",
                "other_ids": {
                    "DOI": [
                        "10.18653/v1/D19-1410"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Nils Reimers and Iryna Gurevych. 2019. Sentence- BERT: Sentence embeddings using Siamese BERT- networks. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natu- ral Language Processing (EMNLP-IJCNLP), pages 3982-3992, Hong Kong, China. Association for Computational Linguistics.",
                "links": null
            },
            "BIBREF4": {
                "ref_id": "b4",
                "title": "A comparison of unsupervised methods for ad hoc cross-lingual document retrieval",
                "authors": [
                    {
                        "first": "Elaine",
                        "middle": [],
                        "last": "Zosa",
                        "suffix": ""
                    },
                    {
                        "first": "Mark",
                        "middle": [],
                        "last": "Granroth-Wilding",
                        "suffix": ""
                    },
                    {
                        "first": "Lidia",
                        "middle": [],
                        "last": "Pivovarova",
                        "suffix": ""
                    }
                ],
                "year": 2020,
                "venue": "Proceedings of the workshop on Cross-Language Search and Summarization of Text and Speech (CLSSTS2020)",
                "volume": "",
                "issue": "",
                "pages": "32--37",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Elaine Zosa, Mark Granroth-Wilding, and Lidia Pivo- varova. 2020. A comparison of unsupervised meth- ods for ad hoc cross-lingual document retrieval. In Proceedings of the workshop on Cross-Language Search and Summarization of Text and Speech (CLSSTS2020), pages 32-37, Marseille, France. Eu- ropean Language Resources Association.",
                "links": null
            }
        },
        "ref_entries": {
            "FIGREF1": {
                "type_str": "figure",
                "num": null,
                "text": "MAP 2D projection of the Latvian data, where black crosses mark query docs (GS) represent gold standard, i.e. manually identified Latvian news of interest to Estonian readers, gray dots represent automatically identified Latvian seeds (identified by string-based search in Estonian and cross-lingual linking to Latvian) and dark-gray dots represent all other Latvian documents.",
                "uris": null
            }
        }
    }
}