What Paper Alerts Are
A paper alert is a saved search query that runs automatically on a schedule.
When papers published since your last alert run match your query, you receive
an email digest containing the matching papers — title, author, category,
abstract excerpt, and a link to the full paper page.
Unlike keyword notifications in social media or news platforms — where the
matching algorithm is opaque and optimised for engagement — paper alerts on
Expertini Research use the same transparent Elasticsearch query infrastructure
as the live search. The same fields are searched, the same boost weights applied.
There is no separate, undisclosed ranking model for alerts.
An alert is nothing more than a scheduled search. If your keyword returns papers
in live search, it returns the same papers in alert digests. What you see in
search is what you get in your inbox.
Three Types of Alerts
You can set alerts based on three different criteria, and combine them freely:
🔍 Keyword Alert
Matches papers where your keyword appears in the title, abstract, keywords, or category fields —
with the same field boost weights as live search (title ^3, abstract ^2, keywords ^1).
Use specific, domain-appropriate terms for the best results. "machine learning" will
return hundreds of papers; "federated learning differential privacy" will return far fewer
but much more targeted results.
📂 Category Alert
Matches all papers published in a specific research category — AI & Data Science,
Medicine & Health, Economics & Finance, etc. This is a keyword-free exact match
on the category field. Use a category alert when you want comprehensive coverage
of a discipline rather than targeted coverage of a specific topic.
You can combine a category alert with a keyword alert for fine-grained filtering.
👤 Author Alert
Matches papers where a specific author name appears in the corresponding_author
or co_authors fields. Author name matching uses the standard analyser — it is
case-insensitive and handles minor variations. For common names, combine with
a category alert to reduce false positives.
The Alert Pipeline — What Happens Each Run
The alert system runs as a scheduled cron job. Here is the full sequence of
operations on each run:
1
Fetch active alerts
The cron job queries the alerts index for all active alert records where the frequency matches the current run (daily or weekly) and the last_triggered timestamp is older than the frequency window.
2
Build time-bounded paper query
For each alert, we query the paper index for documents with status=published AND published_date >= last_triggered. This ensures each run only considers papers published since the previous alert delivery — no duplicates.
3
Apply the match query
The alert's keyword is run through the same multi_match query as live search — across title (^3), abstract (^2), keywords (^1), category (^1). Category and author alerts use term/match queries against their respective fields instead.
4
Score and sort
Matching papers are sorted by relevance score descending, then by published_date descending as a tiebreaker. The top 20 matching papers per alert are included in the digest (if more than 20 match, we include the most relevant).
5
Compose digest email
For each user with matching alerts, we compile a single digest email containing all matching papers grouped by alert. One email per user per run — not one email per alert. This prevents inbox flooding when multiple alerts fire simultaneously.
6
Deliver and update
The email is sent via our SMTP service. After successful delivery, the alert's last_triggered timestamp is updated to the current time. If no papers matched any of a user's alerts, no email is sent for that run.
The core Elasticsearch query for a keyword alert looks like this:
# Alert query — papers since last run matching keyword
{
"query": {
"bool": {
"must": [
{
"multi_match": {
"query": "[alert keyword]",
"fields": ["title^3", "abstract^2", "keywords^1", "category^1"]
}
}
],
"filter": [
{ "term": { "status": "published" } },
{ "range": { "published_date": { "gte": "[last_triggered]" } } }
]
}
},
"sort": [{ "_score": { "order": "desc" } }, { "published_date": { "order": "desc" } }],
"size": 20
}
Delivery Schedule
📅
Daily Digest
Runs every day at 07:00 UTC. Covers papers published in the past 24 hours. Sent only if papers match.
📆
Weekly Digest
Runs every Monday at 07:00 UTC. Covers papers published in the past 7 days. Sent only if papers match.
🔔
Researcher Digest
Researchers receive a separate digest at 08:00 UTC covering comments and citations on their own papers.
Important: if no papers published since your last alert run match
your keyword, no email is sent. A quiet inbox means nothing matched — not that
the system failed. For broad categories like "AI" or "machine learning", daily
alerts will almost always fire. For niche keywords, weekly is often more appropriate.
Setting Up Your Alerts
Alerts are available to both researcher and subscriber accounts. Go to
Paper Alerts
from your dashboard. For each alert:
1
Enter your keyword or select a category — be as specific as your needs require.
2
Choose daily or weekly frequency. Start weekly for niche topics; switch to daily if you find weekly results too infrequent.
3
Click "Add Alert". The alert takes effect immediately — the first run will cover papers published since the moment you created it.
4
Manage, pause, or delete alerts at any time from the same page. You can have up to 10 active alerts on the free plan.
Best Practices for High-Quality Alerts
The quality of your alert results depends almost entirely on the quality of your
keyword. Here is what works and what does not:
✅
Use specific multi-word phrases
"cybersecurity investment optimisation" returns far more targeted results than "cybersecurity" — and you will not miss relevant papers, because specific phrases appear in the abstract of relevant work.
✅
Use the vocabulary of your field
Researchers in your area use consistent terminology. If the field says "large language models", set that alert — not "AI chatbots". The Keyword Suggester tool can help identify the right vocabulary from your own abstract.
✅
Set category + keyword combinations
If you want all AI papers AND papers on bias specifically, set two alerts: one for the category "artificial-intelligence-and-data-science" and one for the keyword "algorithmic bias". You receive both streams separately.
⚠️
Avoid single very common words
"learning" or "data" as a standalone keyword will match hundreds of unrelated papers daily. Always use it as part of a phrase: "federated learning", "synthetic data generation".
⚠️
Avoid brand names or proper nouns alone
"Google" or "OpenAI" as alert keywords will surface papers that merely mention these organisations in passing. Add a topic: "Google BERT language model" is a more useful alert.
Never miss a paper in your field.
Set up keyword, category, and author alerts — free for all registered users.
How Paper Alerts Work
Research
