Relevance scoring mechanism for multiple parameters I have a program which build few attributes those decide relevance between two objects.
attributes are $a_1, a_2, a_3$
Now what are different weighing or scoring mechanism to accumulate all three together.
i.e:

$score = w_1\cdot a_1 + w_2 \cdot a_2 + w_3\cdot a_3$

Deciding weightage parameters.
 A: If I understood your question, you want to build the score based on a weighted average. The answer depends on the relative weight or scoring of each attribute, but that is something you need to decide. Basically each weight is the relative importance of the attribute compared with the rest of attributes, right?
In case of having the same relative weight, each $w_i=\frac{1}{3}$ is obtained from the direct average of the three attributes, so $$score=\frac{a_1+a_2+a_3}{3}=\frac{a_1}{3}+\frac{a_2}{3}+\frac{a_3}{3}=\frac{1}{3}a_1+\frac{1}{3}a_2+\frac{1}{3}a_3$$
The decision of the weights is yours, the only basic rule to modify each weight $w_i$ would be that $w_1+w_2+w_3=1$.
UPDATE: after our conversation, there is another option, which is supervised learning, for instance use a perceptron algorithm to calculate the weights. The perceptron basically is feeded with information about good samples and bad samples (I think in your case would be about the relevance between those two objects), and it calculates the weights dynamically on each feeding cycle, recalculating on each cycle the appropriated values for the weights. After some repetitions of the input-output cycle, the vector of weights should be enough accurate for you needs. My explanation is very generic, but if you are looking for alternatives to the manual decision of the weights, I would suggest you supervised learning. 
