03 / RESEARCH ▸ scroll 000%

RESEARCH ·ASONAM ’17

SODM & SOS

偵測並重排「社交過載」貼文

Detect and rearrange social-overloaded posts to prevent social overload: label boards, learn document-level embeddings, and reorder feeds so negativity doesn't pile up.

DOI ↗ interactive — scroll to step through the method
▸

Background

What “social overload” means, and what the two components do.

Social overload is the stress users feel when a feed piles on too much heavy or negative content at once. This work has two parts: SODM detects socially-overloaded posts, and SOS rearranges the feed so overload doesn't accumulate.

The detector learns document-level embeddings from labelled discussion boards to recognize overloaded posts, and the rearrangement step reorders what a user sees so negativity is spread out rather than stacked. It's an early (2017) example of treating user wellbeing — not just engagement — as something a feed should optimize for.

▸ key terms

Social overload — feed-induced stress from too much heavy content. Embedding — a numeric representation of a piece of text's meaning. Document-level — classifying a whole post, not individual words. Feed rearrangement — reordering the items shown to a user.

SYSTEM · detect → rearrange
first-in-first-out feed reader load support-seeking posts cluster → reader load spikes the “cost of caring” — stress, depression label boards · learn embeddings Prozac Hate Others Word2vec · Skip-Gram overload = {Prozac, Hate} · normal = others balanced sampling avoids board imbalance SODM · CKDGNN detector wordembed CNNfilters K-maxpool(doc) GRNNseq soft-max 95.15% overload-detection accuracy · 5-fold CV beats DCNN · CNN-GRNN · LSTM-GRNN score every post · threshold 0.5 θ = 0.5 0.18 0.91 0.78 0.32 0.83 0.27 posts above 0.5 → flagged social-overload before after · SOS max 3 consecutive overload → insert a calm post reader load stays under the tolerance line 95.15% social-overload detection accuracy 75% of participants reported reduced stress detection + rearrangement = a calmer feed
A first-in-first-out feed lets support-seeking posts cluster — overwhelming the reader.
STEP 00 · the problem

Feeds let negativity pile up

Social feeds display posts first-in-first-out. When many support-seeking, negative posts land together, readers absorb them all at once.

Psychologists call it the cost of caring — repeated exposure drives stress and depression.

STEP 01 · the data

Label boards, learn embeddings

Posts crawled from a BBS are labelled by source: the Prozac and Hate boards as social-overload, others as normal — balanced to avoid skew.

Words are embedded with Word2vec Skip-Gram.

STEP 02 · SODM

A document-level detector

CKDGNN stacks word embeddings → CNN filters → K-max pooling at the document level → a GRNN → softmax, scoring each post's overload probability.

It reaches 95.15% accuracy, beating DCNN, CNN-GRNN and LSTM-GRNN.

STEP 03 · scoring

Flag with a threshold

Every post gets a probability. Anything above the 0.5 threshold is flagged as social-overload — the rest are normal load.

STEP 04 · SOS

Rearrange to protect the reader

The Social-Overload prevention System re-sorts the feed so no more than three overload posts appear in a row — inserting a calmer post to break the streak.

Reader load stays under the tolerance line.

STEP 05 · the result

A measurably calmer feed

Detection hits 95.15% accuracy, and after rearrangement 75% of participants reported that social-overload stress was reduced.