Calculation Results
Enter data and calculateItem-Level Calculation
| Item | User A | User B | Normalised A | Normalised B | Weight | Difference | Product | Contribution | Used |
|---|
Rating Comparison
Nearest-User Ranking
Top Similar Users
| Rank | User | Score | Adjusted score | Common items | Status |
|---|
Step-by-Step Calculation
Formula Used
How to Use
Paste a rating matrix with users in rows. Keep item names in the first row. Leave unknown ratings blank.
Select two users and a similarity method. Choose normalisation and missing-value rules. Then press the calculation button.
Review rankings, item contributions, charts, and predictions. Export the results when finished. Use overlap warnings before trusting conclusions.
Example Data
| User | Item A | Item B | Item C | Item D | Item E |
|---|---|---|---|---|---|
| Alice | 5 | 3 | 4 | Missing | 2 |
| Bob | 4 | 2 | 5 | 1 | Missing |
| Carol | 1 | 5 | 2 | 4 | 5 |
Frequently Asked Questions
What does a similarity score mean?
Higher scores indicate more similar user behaviour. Negative Pearson values indicate opposite rating patterns. Interpretation depends on the selected metric.
Which method works best for ratings?
Pearson often suits rating preferences. Cosine works well for vector direction. Test several methods with validation data.
Why are common items important?
Similarity needs comparable evidence. Very small overlaps create unstable estimates. Use significance weighting and minimum-overlap controls.
What is mean centring?
Mean centring removes each user’s average rating level. It compares preference patterns instead. This helps with generous or strict raters.
How are missing ratings handled?
You can ignore them or apply an imputation rule. Ignoring preserves observed evidence. Imputation enables complete vector comparisons.
What is similarity shrinkage?
Shrinkage reduces scores supported by limited overlap. Larger strengths create stronger penalties. It can improve recommendation stability.
What is inverse user frequency?
It gives more importance to uncommon item interactions. Popular items receive less influence. The idea resembles inverse document frequency.
Can this calculator predict ratings?
Yes. It combines neighbour ratings with similarity weights. Mean-centred prediction also adjusts for rating tendencies.
Why can Pearson be undefined?
Pearson needs variation in both compared vectors. Constant ratings create zero variance. The calculator reports this condition.
Can I use clicks instead of ratings?
Yes. Choose binary normalisation for interaction data. Jaccard and cosine are common choices for implicit feedback.