Automated Content Analysis of Discussion Transcripts
1 like1,001 views
The document discusses the automated analysis of discussion transcripts within asynchronous online education, emphasizing the need for improved, automated content analysis techniques to enhance educational practices. It outlines the Community of Inquiry (CoI) model, which involves cognitive, social, and teaching presences, and presents challenges faced in manually analyzing cognitive presence due to labor intensity and the need for expert coders. The authors explore the use of text mining and machine learning methods to automate the analysis process, highlighting the importance of considering message context and future avenues for research.
Automated Content Analysis of Discussion Transcripts
- 1. Automated Content Analysis of Discussion Transcripts Vitomir Kovanovi´c Dragan Gaˇsevi´c v.kovanovic@ed.ac.uk dgasevic@acm.org School of Informatics, University of Edinburgh Edinburgh, United Kingdom v.kovanovic@ed.ac.uk 31 Aug 2015, University of Edinburgh, United Kingdom
- 2. Asynchronous online discussions - “gold mine of information” (Henri, 1992) • They are frequently used for all types of education delivery, • Their use produced large amount of data about learning processes, • Their use is well supported by the social-constructivist pedagogies. V. Kovanovi´c et al. (EDI) Automated Analysis of Discussion Transcripts 31 Aug 2015 1 / 18
- 3. Asynchronous online discussions - issues and challenges • Produced data is used mainly for research after the courses are over, • Content analysis techniques are complex and time consuming, • Content analysis had almost no impact on educational practice (Donnelly and Gardner, 2011), • There is a need for more proactive use of the data through automation: • Few attempts for automated content analysis, • Focus mostly on surface level characteristics, and • Not based on well established theories of education. V. Kovanovi´c et al. (EDI) Automated Analysis of Discussion Transcripts 31 Aug 2015 2 / 18
- 4. Overall idea Overall idea To examine how we can use text mining for automation of content analysis of discussion transcripts. More specifically, We looked at the automation of content analysis of cognitive presence, one of the three main components of Community of Inquiry framework. V. Kovanovi´c et al. (EDI) Automated Analysis of Discussion Transcripts 31 Aug 2015 3 / 18
- 5. Community of Inquiry (CoI) model Community of Inquiry model (Garrison, Anderson, and Archer, 1999) Conceptual framework outlying important constructs that define worthwhile educational experience in distance education setting. Three presences: • Social presence: relationships and social climate in a community. • Cognitive presence: phases of cognitive engagement and knowledge construction. • Teaching presence: instructional role during social learning. CoI model is: • Extensively researched and validated. • Adopts Content Analysis for assessment of presences. V. Kovanovi´c et al. (EDI) Automated Analysis of Discussion Transcripts 31 Aug 2015 4 / 18
- 6. Community of Inquiry (CoI) model Community of Inquiry model (Garrison, Anderson, and Archer, 1999) Conceptual framework outlying important constructs that define worthwhile educational experience in distance education setting. Three presences: • Social presence: relationships and social climate in a community. • Cognitive presence: phases of cognitive engagement and knowledge construction. • Teaching presence: instructional role during social learning. CoI model is: • Extensively researched and validated. • Adopts Content Analysis for assessment of presences. V. Kovanovi´c et al. (EDI) Automated Analysis of Discussion Transcripts 31 Aug 2015 4 / 18
- 7. Cognitive presence Cognitive Presence “an extent to which the participants in any particular configuration of a community of inquiry are able to construct meaning through sustained communication.” (Garrison, Anderson, and Archer, 1999, p .89) Four phases of cognitive presence: 1 Triggering event: Some issue, dilemma or problem is identified. 2 Exploration: Students move between private world of reflection and shared world of social knowledge construction. 3 Integration: Students filter irrelevant information and synthesize new knowledge. 4 Resolution: Students analyze practical applicability, test different hypotheses, and start a new learning cycle. V. Kovanovi´c et al. (EDI) Automated Analysis of Discussion Transcripts 31 Aug 2015 5 / 18
- 8. Cognitive presence coding scheme • Use of whole message as unit of analysis, • Look for particular indicators of different sociocognitive processes, • Requires expertise with coding instrument and domain knowledge. V. Kovanovi´c et al. (EDI) Automated Analysis of Discussion Transcripts 31 Aug 2015 6 / 18
- 9. Community of Inquiry (CoI) model Issues and challenges: • Very labor intensive, • Crude coding scheme, • Requires experienced coders, • Can’t be used for real-time monitoring, • Not explaining reasons behind observed levels of presences, and • Not providing suggestions and guidelines for instructors to direct their pedagogical decisions. V. Kovanovi´c et al. (EDI) Automated Analysis of Discussion Transcripts 31 Aug 2015 7 / 18
- 10. Data set • Six offerings of graduate level course in software engineering. • Total of 1747 messages, 81 students, • Manually coded by two coders (agreement = 98.1%, Cohen’s κ = 0.974), ID Phase Messages (%) 0 Other 140 8.01% 1 Triggering Event 308 17.63% 2 Exploration 684 39.17% 3 Integration 508 29.08% 4 Resolution 107 6.12% All phases 1747 100% Number of Messages in Different Phases of Cognitive Presence V. Kovanovi´c et al. (EDI) Automated Analysis of Discussion Transcripts 31 Aug 2015 8 / 18
- 11. Feature extraction • Unigrams, Bigrams and Trigrams, • Part-of-Speech Bigrams and Trigrams, • Backoff Bigrams and Trigrams: Example: “John is working.” Bigrams: • john is, • is working. Backoff Bigrams: • john verb , • noun is, • is verb • verb working. V. Kovanovi´c et al. (EDI) Automated Analysis of Discussion Transcripts 31 Aug 2015 9 / 18
- 12. Feature extraction • Dependency triplets: rel, head, modifier Example: “Bills on ports and immigration were submitted by Senator Brownback, Republican of Kansas.” nsubjpass, submitted, Bills auxpass, submitted, were agent, submitted, Brownback nn, Brownback, Senator appos, Brownback, Republican prep of, Republican, Kansas prep on, Bills, ports conj and, ports, immigration prep on, Bills, immigration V. Kovanovi´c et al. (EDI) Automated Analysis of Discussion Transcripts 31 Aug 2015 10 / 18
- 13. Feature extraction • Backoff dependency triplets: Example: “Bills on ports and immigration were submitted by Senator Brownback, Republican of Kansas.” Dependency triplet: • conj and, ports, immigration Backoff dependency triplets: • conj and, noun , immigration • conj and, ports, noun • conj and, noun , noun V. Kovanovi´c et al. (EDI) Automated Analysis of Discussion Transcripts 31 Aug 2015 11 / 18
- 14. Additional features • Number of named entities in the message Brainstorming should involve more concepts than posing a question, • Is message first in the discussion? Posing questions is more likely to be initiating discussions, • Is message a reply to the first message in the discussion? V. Kovanovi´c et al. (EDI) Automated Analysis of Discussion Transcripts 31 Aug 2015 12 / 18
- 15. Classification Classifier: • SVM classifier with RBF kernel. • Accuracy and kernel parameter tuning evaluated using nested 5-fold cross-validation. • Only features with support of 10 or more, • Accuracy evaluated using 10 fold cross-validation, • Comparison of models using McNemar’s test. Implementation: • Implemented in Java, • Feature extraction using Stanford CoreNLP1 toolkit, • Tokenization, Part-of-Speech, and Dependency parsing modules • Classification using Weka (Witten, Frank, and Hall, 2011) and LibSVM (Chang and Lin, 2011), and • Statistical comparison using Java Statistical Classes (JSC)2 1http://nlp.stanford.edu/software/corenlp.shtml 2http://www.jsc.nildram.co.uk/index.htm V. Kovanovi´c et al. (EDI) Automated Analysis of Discussion Transcripts 31 Aug 2015 13 / 18
- 16. Results • We achieved Cohen’s κ of 0.42 for our classification problem. • Better then the existing Neural Network system (Cohen’s κ=0.31). • Unigram baseline model achieved Cohen’s κ of 0.33. Error analysis: Predicted Actual Other Trigg. Expl. Integ. Resol. Other 17 04 05 02 00 Triggering 01 42 ⇒1 14 03 01 Exploration 02 09 98 24 04 Integration 01 03 38 ⇐1,2 56 04 Resolution 00 00 03 15 ⇐2 03 Confusion Matrix V. Kovanovi´c et al. (EDI) Automated Analysis of Discussion Transcripts 31 Aug 2015 14 / 18
- 17. Challenges 1 Effect of the large relative size of the exploration class, 2 Effect of the code-up rule for coding, 3 No relative importance of features, and 4 Context is not taken into the account. Code-up rule for coding V. Kovanovi´c et al. (EDI) Automated Analysis of Discussion Transcripts 31 Aug 2015 15 / 18
- 18. In progress: making use of tread context • Discussions (and students’ learning) progresses from triggering to resolutions. • Content of a message depends on the content of the previous messages. • Content of a message depends on the learning progress of a given student. Model for message classification V. Kovanovi´c et al. (EDI) Automated Analysis of Discussion Transcripts 31 Aug 2015 16 / 18
- 19. Approach: Hidden Markov models (HMMs) & Conditional random fields (CRFs) • Hidden Markov Models: • HMMs used to models system states and their transitions in a variety of contexts. • Widely used, Bayesian Knowledge Tracing models based on HMMs. • Challenges with HMM: • Can this be modeled as HMM (2nd order HMMs?) • Dependency only on a single previous state, • One manifest variable for each state • Conditional random fields: • Used for structured predictions (e.g., speech recognition) • For speech recognition, take into the account the classes of all letters in a word. • Widely used in natural language processing, • More flexible than HMMs, • Challenges with CRF: • Too many parameters to estimate with little data V. Kovanovi´c et al. (EDI) Automated Analysis of Discussion Transcripts 31 Aug 2015 17 / 18
- 20. Conclusions and future work Summary: • Promising path to explore, • Use of backoff trigrams, plain and backoff dependency triplets, entity count and first message indicator seems useful, Future work: • Additional types of features which look at the context of previous messages (e.g., convergence vs. divergence), • Moving away from SVM, explore other classification methods which are better at explanation • Give associated probabilities for each classification, • Give relative importance of different features. Challenges: • Challenges with message unit of analysis and surface-level features, • Low frequency of resolution messages. V. Kovanovi´c et al. (EDI) Automated Analysis of Discussion Transcripts 31 Aug 2015 18 / 18
- 22. References I Chang, Chih-Chung and Chih-Jen Lin (2011). “LIBSVM: A library for support vector machines”. In: ACM Transactions on Intelligent Systems and Technology 2 (3), 27:1–27:27. Donnelly, Roisin and John Gardner (2011). “Content analysis of computer conferencing transcripts”. In: Interactive Learning Environments 19.4, pp. 303–315. Garrison, D. Randy, Terry Anderson, and Walter Archer (1999). “Critical Inquiry in a Text-Based Environment: Computer Conferencing in Higher Education”. In: The Internet and Higher Education 2.2–3, pp. 87–105. Henri, France (1992). “Computer Conferencing and Content Analysis”. en. In: Collaborative Learning Through Computer Conferencing, pp. 117–136. Witten, Ian H., Eibe Frank, and Mark A. Hall (2011). Data Mining: Practical Machine Learning Tools and Techniques, Third Edition. 3rd ed. Morgan Kaufmann.