Extract Markov chains from client and therapist categorical labels

This tool comes from a project exploring client-therapist interaction based on categorical labels from moment to moment. The tool extracts transition probability matrix based on discrete time Markov chain from the categorical labels of both client and therapist.

Copy the repository

git clone https://github.com/XinY-Z/markov-chain.git

Install the required packages

install.packages("data.table")
install.packages("stringr")

Data

The data should be in the format of a dataframe (preferrably data.table) with at least the following variables:

• Session ID
• Speaker labels (e.g., T or P)
• Categorical labels (e.g., therapist's interventions, client's themes)

The data should be sorted in chronological order.

Usage

First, customize the options in utils.R to your own data. For example

LABEL <- 'code'           # column name for categorical codes
SESSION_ID <- 'session'   # column name for session ID
SPEAKER <- 'speaker'      # column name for speaker
THERAPIST <- 'T'          # therapist label
CLIENT <- 'P'             # client label
CONCAT <- F               # consider codes that co-occur or not
# By setting CONCAT to TRUE, co-occurred codes are considered as a single code, 
# e.g., 'reflection,socratic' is considered one code instead of two separate codes.

Second, customize the options in getMarkov.R to your own data. For example

INPUT_PATH <- <your data file path>
OUTPUT_PATH <- <your output file path>
SESSION_ID <- 'session'   # column name for session ID
TOPIC <- 'REC'            # focus on code REC
CONCAT <- F               # consider therapist codes that co-occur or not
STARTWITH <- 'P'          # consider only transition from client to therapist
# STARTWITH determines the direction of condition, e.g., STARTWITH=P means looking 
# at P -> T sequences.

Finally, run the script getMarkov.R to get the transition probability matrix.

source('getMarkov.R')
# This will export the processed file to the output file path you specified.

Example result

The first column is the session ID, the following columns are code sequences and their transition probabilities, e.g., a probability of transitioning from MOOD code to ADW code in session 8510 is 0.00311.

session     MOOD2ADP     MOOD2ADW     MOOD2AF     MOOD2CO      ...
   8430      0.00000      0.00000     0.02272     0.00000      ...
   8433      0.00000      0.00000     0.00000     0.00000      ...
   8472      0.00000      0.00000     0.00000     0.00000      ...
   8510      0.00000      0.00311     0.02180     0.00000      ...
    ...          ...          ...         ...         ...      ...

License

This project is licensed under the MIT License - see the LICENSE file for details.