algorithm: the input of the whole model includes four parts: the selected discourse D,
the context C, the joint distribution of discourse and context under the given parameter
conditions
, and the conditional distribution of context under the given
parameter and discourse conditions
, and the output is the model
parameters . The model first chooses an initial value of for the parameter and
starts iterating, producing an estimate for at each iteration, denoted , where i is the
number of rounds of iteration. At the th iteration, the expectation is computed:
(1)
Then the parameter , which makes Eq. (1) maximized, determines the estimate of
the parameter for the nd iteration :
(2)
The process of Eq. (1) and Eq. (2) is repeated continuously until convergence, so
that the context parameters can be approximated and thus the discourse can be
judged to belong to which context distribution [24].
4.2. DISCUSSION OF CONTEXTUAL FORMALIZATION ISSUES
AND TIME FACTORS
In order to reflect the dynamics of the model, it is also necessary to take the time
factor into account, adding the stochastic process model that fits the nature of
compliance is an effective method.
1.
The complexity of the context and the formalization of ideas, the current
difficulties in the application of the model lies in the formal description of the
discourse and the context, here you can refer to the relevant theories in formal
pragmatics. In addition, two ideas about the description of context and
discourse can be referred to here. First, consider the factors affecting the
context as variables, and because of too many influencing factors you can set
the number of variables to be infinite, and build the model through limit theory
or infinite series. The second is not to consider each factor affecting the
context, but to abstract the common characteristics of each factor and build the
model in the way of structure and variety, similar to the construction of group,
ring and domain in abstract algebra.
2. If we consider the time factor in the interaction process between the two parties
using the discourse, the model can be constructed by using the Hidden Markov
Chain. In this case, the state sequence of the potential Markov chain is the
context state , which is unobservable, while the observation value is
the discourse obtained after selection, and its transfer probability at
the n rd state is:
i
C
i
i
i+1
i
https://doi.org/10.17993/3cemp.2024.130153.176-194
183
3C Empresa. Investigación y pensamiento crítico. ISSN: 2254-3376
Ed. 53 Iss.13 N.1 January - March, 2024