Hivemind: Custom Trends

Hivemind custom trends allow users to define their own concepts and narratives and measure them consistently through time. By specifying keywords, phrases, or semantic descriptions, users can create bespoke trend signals that reflect portfolio-specific risks, themes, or investment theses.

Custom trends are point-in-time and constructed using the same underlying Hivemind ingestion and scoring framework as system-defined trends.

UI Access

Custom trends can be created and explored in the ForecastOS UI, where users define concepts, review historical behavior, and compare custom narratives against existing Hivemind trends.

Hivemind Custom Trends

API Access

Custom trends are available through the ForecastOS API for automated creation, updating, and retrieval.

Example: Fetching Custom Trend Popularity Evolution via API

curl -X POST "https://app.forecastos.com/api/v1/trends/custom" \
-H "Authorization: Bearer YOUR_API_TOKEN" \
-H "Content-Type: application/json" \
-d '{
  "trend": {
    "text": "artificial intelligence",
    "sensitivity": "medium",
    "start_date": "2020-01-01"
  }
}'

Query Parameters

Parameter Type Default Required Description
text string - Yes The trend to search for. Use noun-like terms such as a person's name (Elon Musk), objects (electric cars), or general topics (artificial intelligence).
sensitivity string "medium" No Controls how trend text is matched to relevant mentions. Semantic similarity options are "low", "medium", "high", while text similarity options are "exact" and "fuzzy".
start_date string "2015-01-01" No Optional start date (YYYY-MM-DD) for the trend calculation. Defaults to 2015-01-01 if not provided.

Response

{
  "rolling_90d_popularity": {
    "2026-01-01": "150.0",
    "2026-01-02": "154.0",
    "2026-01-03": "158.0",
    "2026-01-04": "162.0",
    "2026-01-05": "166.0",
    "2026-01-06": "170.0"
  },
  "rolling_365d_popularity": {
    "2026-01-01": "100.0",
    "2026-01-02": "108.0",
    "2026-01-03": "116.0",
    "2026-01-04": "124.0",
    "2026-01-05": "132.0",
    "2026-01-06": "140.0"
  }
}

Open-Source Access

The open-source ForecastOS Python library provides helpers for submitting custom trend definitions and consuming the resulting time series in research workflows.

Concept embedding, scoring, and source processing are handled within the managed Hivemind platform.

Example: Fetching Custom Trend Popularity Evolution via OS

import forecastos as fos

df_custom_trend = fos.CustomTrend.get_df(
  text='artificial intelligence', 
  sensitivity='medium', 
  start_date='2020-01-01'
)

Parameters

Parameter Type Default Required Description
text string - Yes The trend to search for. Use noun-like terms such as a person's name (Elon Musk), objects (electric cars), or general topics (artificial intelligence).
sensitivity string "medium" No Controls how trend text is matched to relevant mentions. Semantic similarity options are "low", "medium", "high", while text similarity options are "exact" and "fuzzy".
start_date string "2015-01-01" No Optional start date (YYYY-MM-DD) for the trend calculation. Defaults to 2015-01-01 if not provided.

This returns a time-series DataFrame for the popularity evolution of the trend.

Next: Hivemind Company Exposures

Let's explore Hivemind Company Exposures next.