To be precise, EI (Compendex) does not set this rule. EI is an indexing database. The rule is set by the publisher of the conference, which in this case is IEEE.

There is no single, official "magic number" (e.g., 20%) that IEEE states for all conferences. The decision is not just a percentage, but a detailed review by the conference editors.

However, based on the standards used by most IEEE conference organizers (Technical Program Chairs and Editors) who use the iThenticate/CrossCheck tool, here are the general guidelines.

General Guidelines for Similarity Rates (IEEE Conferences)

Editors look at two different numbers in the iThenticate report:

1. Overall Similarity Index (The Total Percentage)

   • < 20%: This is generally considered a safe score and will almost always pass the initial screening.

   • 20% - 30%: This is an acceptable range. An editor will likely review the report in detail to see where the similarity is coming from (see "What Editors Look For" below).

   • > 30% - 40%: This is a high-risk score. The paper will be heavily scrutinized. If the overlap is in the core "results" or "conclusion" sections, it may be rejected.

   • > 40%: This score will most likely result in an immediate desk rejection by the conference organizers.

2. Single Source Similarity (The More Important Metric)

   • This is the percentage of overlap with a single source (e.g., one specific journal article, website, or your own previous paper).

   • < 5%: This is the ideal target for any single source.

   • > 10%: This is a major red flag. If your paper matches more than 10% with a single existing paper (even your own), it suggests significant plagiarism or "self-plagiarism" (text recycling), which IEEE strictly prohibits.

Why the Percentage Isn't the Whole Story

A high percentage is not always bad, and a low percentage is not always safe. The editor will not just look at the number; they will look at the full report.

• "Good" Overlap (Often Ignored): The editor will usually exclude matches from:

  • Bibliography / References: The reference list will always match other papers.

  • Standard Methods/Formulas: Common technical phrases, mathematical formulas, or descriptions of standard methodologies (e.g., "The dataset was divided into a training and testing set...") are expected to overlap.

  • Author Affiliations: These will match.

• "Bad" Overlap (The "Deal-Breakers"): The editor is looking for:

  • Plagiarism: Copying sentences or paragraphs from another author's paper without citation and quotation marks.

  • Self-Plagiarism (Text Recycling): This is a serious offense in IEEE's view. You cannot copy-paste large sections (like the introduction, conclusion, or main results) from your own previously published paper (e.g., a journal version of a previous conference paper).

Final Recommendation

1. Aim for < 20% Overall: This is the safest target to ensure your paper passes the initial check without issues.

2. Aim for < 5% Single Source: This is the most important rule.

3. Check the Specific Conference: Some top-tier IEEE conferences (e.g., CVPR, ICCV) may have even stricter requirements. Always check the "Author Guidelines" or "Submission" page on the conference's official website.

Would you like a guide on how to properly paraphrase your own work to avoid self-plagiarism?