Repository navigation
[Help]Extracted facts/triples consistently missing temporal information (timestamps/dates) #987
Description
As shown in the attached screenshot, the output only includes the entity/relation data without any temporal context:
Expected Behavior:
I expected the extracted facts to include a timestamp, date, or some form of temporal grounding associated with each fact.
Activity
WaylandYang commentedon Sep 28, 2026
Thanks for reporting this, and for the screenshot. We reproduced it with a report shaped like yours, and it led to real fixes.
What is intended. A fact gets a start and an end only when its own sentence says when. "码表 所属市场为 全球" does not, so it has no interval. We never fill one in from the upload time.
What was wrong on our side.
Where it stands. The design is in ADR 0064 (#989). The first part, a bench that measures this, is merged (#990). Reading time words and section dates is in review (#991), and showing "as of " on the panel follows it. I will post here when they land.
One question, to make sure your case is the same one. Which version are you running (the image tag, or the version shown at the top right)? And if you can share a short sample of one document, even with the content replaced, we can add its shape to the bench.
A statement is attested by the date its own section speaks from, and …
bulexu commentedon Sep 30, 2026
Thank you so much for the prompt and detailed response, as well as the quick fixes! It’s great to see the progress with ADR 0064.
To answer your questions:
Version: I am currently running [v0.1.0-rc6-51-g7cfeedf].
Document Sample: I apologize that I cannot share the original document sample due to data privacy and internal confidentiality policies.
Thanks again for your hard work and support!
WaylandYang commentedon Oct 10, 2026
Thank you again for the report. The fixes it led to are released in v0.1.0-rc8: a time word is read as the statement's time and measured from the date of its own section, and a statement with no time of its own is attested by the date its document speaks from. Closing this as resolved. If a document still loses its dates on rc8, please reopen with the sentence and we will look. Thanks for your help.
bulexu commentedon Oct 16, 2026
As shown in the attached screenshot (Figure 1), following the specifications in the original text (Figure 2), the time attribute should be extracted properly from the input text/time, but the extraction currently fails.
Additionally, I adjusted the chunking logic for Excel files: each row is now treated as a separate chunk, with the table headers prepended to every chunk.
WaylandYang commentedon Oct 16, 2026
Thank you for coming back with the screenshots and the row; that was enough to reproduce it.
We built a sheet with your 16 columns, one row per section with the header repeated (the chunking you described), and created the attributes the way the ontology page does. Two runs before the fix kept 9 and 5 of the 17 values the rows state, and the 处理状态, 优先级 and 需求编号 columns were lost in whole rows at a time. Three things were wrong:
待审核,P1orREQ-0019as separate entities, and in some sections made the sheet name (需求记录) the subject of every cell. A statement whose object is an entity can never fill an attribute.2026-06-23 16:00was sometimes listed as an entity in the same way, with the same result.2026/07/01next to2026-07-01.The fix is in #1252. After it, two runs over the same sheet both kept all 17 values, with 提出时间 and 通知时间 on every row that has them. It is in review and will be in the next release. Documents already extracted keep their current facts; reprocess them after upgrading to pick it up.
Two things would help us confirm your case is the same one:
Thank you again for the detailed report.
WaylandYang commentedon Oct 16, 2026
The fix from #1252 is now in
devand will be in the next release. Documents extracted before it keep their current facts; reprocess them after upgrading. If a row still loses a time column after that, please reopen with the row.