jayeshchoudhari / hmhp

Hidden Markov Hawkes Process - Model for Analyzing Topical Transitions in text based cascades in Social Networks.

HMHP

Hidden Markov Hawkes Process is probabilistic model proposed in Discovering Topical Interactions in Text-based Cascades using Hidden Markov Hawkes Processes

Executing HMHP model code (inside HMHPModel folder)

The cpp file hmhp_EstAll_GroupedWuv.cpp is the code for the HMHP model. To run the model, the input files required are:

• events file
• documents file
• followers map
• indegree and outdegree file for each node
• candidate parents for each event and a file containing exponential time difference value for each candidate parent for each event (this can be precomputed from the events file and followers map):

To compile the code:

make

To execute the code, argv[1] = BURN-IN, argv[2] = Total No. Of Iterations, argv[3] = path to input files, and argv[4] = path to output files. Code can be executed as follows:

./mainHMHP 200 301 inputFilesOurModel.txt outputFilesOurModel.txt

As we record the inferred parent and topic assignments only at every 10th iteration after the BURN-IN period, the total number of iterations must be greater than 10 and greater than BURN-IN period.

The code outputs various files as follows:

• parent assignment (recorded at every 10th iteration after BURN-IN)
• topic assignment (recorded at every 10th iteration after BURN-IN)
• avg parent assignment (probability of candidate parent event)
• grouped wuv values

After the execution of the above code and once all the files are in place, one can execute the script evaluateHMHP.sh

The file format for the input files is as follows:

1. Events File: Each line of events file describes an event with 5 space separated values as follows:

event_time user_node parent_event_id topic_id level_info

2. Documents File: i-th line has a document corresponding to i-th event in the events file. Each document is set of space separated (int) word-ids.

3. Followers Map: Each line of the followers map is as follows (all the values are space separated):

NumOfFollowers userId followerId-1 followerId-2 ... followerId-M

4. Indegree and Outdegree Files: Each line of the both the files has two space separated values:

out/in-degree userId

5. Candidate Parents File: Each line contains the information of the candidate parent events for each event. Each line has space separated values as follows:

NumOfCandParents EventId CandParentId_1 CandParentId_2 ... CandParentId_100 

6. Candidate Parents Exponential Time-difference file: Each line contains the information of the exponential time difference between the candidate current event and the candidate parent event. Each line has space separated values as follows:

NumOfCandParents EventId ExpTimeDiffWithCandParentId_1 ExpTimeDiffWithCandParentId_2 ... ExpTimeDiffWithCandParentId_100

To get the candidate parent files, run the code "getStoreTopKCandParents.cpp". The program takes the events file as input and outputs two files -- one for the candidate parents and other is the exponential time difference file. The program can be run as follows:

g++ -std=c++11 -Wall -O2 -g -o getTop100CandParents  getStoreTopKCandParents.cpp

./getTop100CandParents pathToEventsFile pathToCandidateParentFile pathToCandidateParentExpTimeDiffFile