Meeting - 06/08

[amritbhanu]

Stack Results.

New Citemap Results

20

Paper Related graphs

Evaluation criteria

• The score remains constant till 200, it goes up when we do 300 and stays constant or get down.

  Frequency of parameters per max score

• Pits A dataset. For each iteration i, with each label, its max score is selected. And these are the no of set of parameters which produced that stability score. Just for 2 labels.

Boxplot graph to show how parameter ranges are suffered.

[amritbhanu]

RQ1: Does tuning improve the stability scores of LDA?

• Show it with a table.

RQ2: Does different data need different configurations to make LDA stable?

• Parameter ranges, boxplot with different datasets

RQ3: Does tuning changes some predefined notions of topic modelling?

• Like topic sizes with citemap dataset.

RQ5: Is tuning impractically slow?

• No of evaluations

RQ6: Does terms remain constant in a topic over different runs?

• heatmap

[amritbhanu]

RQ 1: Is it a quirk of the implementation?

Explain how the likelihood is measured in LDA?

Algorithm = untuned

Stability vs evaluation = let see for stack citemap

Stats on tha AUC