---

tags: cofacts, meeting note GA: UA-98468513-3 —

20220209 會議記錄

:::info

• 所有會議記錄
• 線上出席：orz, bil, nonumpa
• https://gather.town/app/z3x18KQFgZCX8MeZ/cofacts :::

:potable_water: Release pipeline

:star: Released to production

:globe_with_meridians: Site

Cofacts user blocking mechanism 已全數完成。

• https://github.com/cofacts/rumors-site/releases/tag/release%2F20220126
• https://github.com/cofacts/rumors-site/releases/tag/release%2F20220127

檢舉表單

• Spreadsheet: https://docs.google.com/spreadsheets/d/1q-iMJ9tZtoyHNfwInAb3z-eIg2x4-w1yp7XGTiaLB-w/edit
• Public link: https://docs.google.com/spreadsheets/d/e/2PACX-1vRdcwXdC36xfgXfSMSk527Zbel9A-__vwRXkQ0NjkzSXoSPETCFc7sI7SoaAFdPCfskugtQL-Md8JgH/pubhtml?gid=438362561&single=true
• Markdown 格式：https://github.com/cofacts/takedowns/blob/master/2022/0129-spammers-and-2nd-scammers.md

:eye: Under review

• Multimedia support phase 0

圖片 hash 效果

:::info Previous research: https://g0v.hackmd.io/bhL6csQ8T1e81E2De7ZS5Q#Search-with-similarity 針對「如何 search vector 間的距離」 本週探討 hash 作為 vector 本身的可行性，以及找尋適合用於 index images & dedup 的 hash 參數 :::

• UI visualization: https://github.com/MrOrz/cofacts-collected-image
• Cofacts research on image-hash
• Current parameter (bits=16) is good for dedup
  • TBA: analysis of bits=12 (144-bit) hash
• Similarity search
  • also use bits=16?
  • d<=10 –> Highly similar (95% similar)
• 如何找 d <= 10 的 hash? [name=nonumpa]
  • 我們的數量級 10K images，直接掃 [name=mrorz]
    • 256 bit hash 每 4 bit 做 XOR 查表 –> 64 次 XOR 與加法可完成一個 hamming distance 計算
    • 改成 Uint16Array 每 16 bit 做 XOR 查表 –> 16 次 XOR 與加法（但 dist lookup table 要 65536 個這麼大）