Audience rating predication before broadcasting based on context case-based reasoning

In the early 1990s, China's television industry began to change from "integration of production and broadcasting" to "separation of production and broadcasting". As broadcast institutions, Television stations need to buy the right or copyright of television programs from production companies. The accurate audience rating prediction of the pre-broadcast ratings of TV programs can help TV stations to make reasonable purchases of TV programs, reducing investment risks and operating risks of TV programs. At the same time, the audience rating prediction of pre-broadcast ratings can also provide a basis for advertisers to formulate advertising strategies. Therefore, it is of great significance to study the audience rating prediction of TV ratings before broadcasting. This paper studies the pre-broadcast audience ratings prediction method based on situational case reasoning, makes full use of the accumulated historical ratings data, and uses the matching and reuse of historical TV program cases and new program cases to predict the pre-broadcast audience ratings of new programs to make up the shortcomings of high cost and low efficiency for the traditional rating prediction method, and provide new ideas for the prediction of pre-broadcast ratings of TV programs. The main research work of this paper includes the following three aspects: 1) In view of the strong structural characteristics of TV programs, the framework method is used to construct an effective and reasonable case representation for TV programs, which is conducive to the further expansion of case knowledge. At the same time, introducing psychology and the context of knowledge managements a specific form of expression of the case, we build a multi-level context structure of TV drama case expression that blends internal and external context; 2) In the case retrieval, firstly the distance between the target case and the source case is calculated, the distance between the two cases according to the weight of each situation is then calculated, so as to obtain the similarity between the two cases. For the multi-valued context that exists in the TV program case expression, we construct a local similarity measurement method for the multi-value symbolic context, that is, multi-value matching strategy; 3) Case reuse is the difficult part in case-based reasoning, which is mainly attributed to its domain dependence. For the case matching of the target case and similar cases in case reuse, we construct a context coefficient adjustment rule based on the difference context, that is, according to the context matching of the target case and the most similar case retrieved, the solution of the difference context is used to adjust the solution of the most similar case to obtain the proposed solution of the target case. Finally, based on the 8-month audience rating data of East China region, this paper uses TV drama ratings as a specific case for experimental analysis. The results show that when calculating the similarity between the target case and the source case in the case retrieval phase, for the calculation of similarity degree of the multi-valued symbolic situation, the recall rate and the precision rate with using the multi-value matching strategy are higher than those without using the strategy; when the adjustment rules are adopted, the prediction accuracy rate of the ratings is higher than that without such rules, and the proportion of accurate cases and deviation cases in the prediction results is higher; from the overall performance test, it can be seen that the pre-broadcast audience ratings prediction model proposed in this paper has achieved good prediction results and demonstrated its effectiveness and rationality. Predicting the ratings of TV programs based on case-base reasoning and adopting a data-driven strategy to predict the broadcasts not only provides a new way to effectively predict the ratings of TV programs, but also it has certain reference significance for predicting digital programs such as online videos and movies. Subsequent work will further study the feature selection of case-base construction and case retrieval strategy optimization to obtain better pre-broadcast prediction performance; in the future, we can further increase the sample size, extend the time span, enrich data features, and conduct larger scale of ratings trend research.