News topic tracking method

Abstract

The invention relates to a news topic tracking method, which combines a dynamic topic model with a BP neural network, and comprises the following steps of: changing the corresponding weight of a time-event space model according to actual needs; completing a mapping relation between a sample and a result through BP neural network training, dividing the test corpus into a plurality of parts, dynamically training a topic model, continuously updating keywords in the model and weights of the keywords, retraining a neural network, and performing new classification to complete event tracking. According to the method, a self-adaptive updating strategy is introduced into the dynamic topic model, the topic model is continuously updated, the problem that the correct rate is reduced due to topic offset in the event tracking process is solved, a good topic tracking effect can be achieved, and the requirements of practical application can be well met.

Description

technical field [0001] The invention belongs to the technical field of text information processing, and in particular relates to a news topic tracking method. Background technique [0002] In the topic tracking task, it is stipulated that only a small number of reports are used to train the topic model. Insufficient data makes it difficult for the constructed topic model to fully express the topic. Among the existing topic tracking technologies, there are many algorithms based on static topic models. Static topic model, as the name suggests, is a static model. Its design idea is to give few training reports related to news report topics, convert the reports into models after keyword extraction, and then perform the same preprocessing on the test corpus , get the corresponding model, and finally judge whether the model is related to the training model according to the similarity between the model to be tested and the training model, that is, judge whether the topic is an old...

AI Technical Summary

Problems solved by technology

[0004] The KNN algorithm in the prior art is an instance-based classification algorithm. It does not require complex calculations, nor does it require prior knowledge of the corpus to train the model. It only needs to pass the vote by obtaining the K nearest neighbor sample points closest to the test point. To determine the attribution of test points, the KNN algorithm relies on a very small number of sample points to determine the attribution of test points. If the distribution of sample points in the data is unbalanced, the existence of a certain type of sample points with a large proportion will lead to an increase in the false detection rate.

Moreover, with the increase of news data, the impact of topic drift is also a problem that KNN cannot effectively solve.

[0005] In the existing topic tracking technology, the topic tracking method based on the static topic model does not incorporate an adaptive update mechanism, resulting in the system not being able to update in real time and not being able to solve the problem of topic drifting well

Images