taki0112 / vector_similarity

Python, Java implementation of TS-SS called from "A Hybrid Geometric Approach for Measuring Similarity Level Among Documents and Document Clustering"

Vector_Similarity

• Python, Java implementation of TS-SS called from "A Hybrid Geometric Approach for Measuring Similarity Level Among Documents and Document Clustering"
• Also, I have summarized "A Hybrid Geometric Approach for Measuring Similarity Level Among Documents and Document Clustering"
• I recommend TS-SS instead of Cosine distance or Euclidean distance.

The reasons are...

Cosine drawbacks

Euclidean drawbacks

Triangle's Area Similarity (TS)

Sector's Area Similarity (SS)

TS-SS

Results

Conclusion

• In biggest dataset, TS-SS outperforms Cosine with a significant difference, while in other datasets TS-SS outperforms Cosine slightly

• Therefore, the significant better result of TS-SS in biggest dataset justifies the robustness and reliability of the model for big data and real world data where the variety of documents/texts are high

Reference

[1] A Hybrid Geometric Approach for Measuring Similarity Level Among Documents and Document Clustering [link1] [link2] [View Article]

vector_similarity's Issues

angle Theta converted 2x to radian

Hi,

thanks a lot for the great implementation of TS-SS!
I noticed that in your as well as all other implementations I have found, the angle theta is converted to radians, although it already is in radian.

See the following line here:
https://github.com/taki0112/Vector_Similarity/blob/3a1f6248aba6be8eed865339a26e800dcf02c028/python/TS_SS/Vector_Similarity.py#L20C1-L20C43

Is there a reason for this conversion? As far as I can tell Theta() outputs the angle in radian (as also demonstrated by the fact that we add math.radian(10) to the acos).

Similarly, the area of the sector is calculated with the formula for the angle in degrees, but again, as far as I can tell Theta is in radian.

Thanks a lot in advance!
Clemens

I just try python version of ts-ss similarity but found we cannot produce similarity in bulk for matrices of v1(m x n) and v2(m x k).

I just try python version of ts-ss similarity but found we cannot produce similarity in bulk for matrices of v1(m x n) and v2(m x k).

It need vectorization to support similarity measure for large matrices in bulk.
Current approch is quite slow.

License

What is the License for this software and the TS algorithms ?
are they open source for commercial use?

Vectorized Calculation?

If we want to calculate for large collections of document, can ts-ss be vectorized like cosine similarity

Why euclidean and ts ss result exceeding 1?

I tried the java version. When I run the example from the code, the result as follows
euclidean: 2.23606797749979
cosine similarity: 0.9999999999999998
ts ss: 4.6395825669994775E-4

I get that the cosine similarity is 0.99, because I expect it to be out of 1 or 100%. But why euclidean and ts ss exceed 100%. How to measure it then?