CN107944007A - Recommend method in a kind of personalized dining room of combination contextual information - Google Patents

Abstract

The invention relates to a personalized restaurant recommendation method combined with context information, which includes establishing a rule base, a cold start phase, and a user data analysis phase; wherein, the established rule base contains short-term preference rules and fixed preference rules. In the cold start phase, use cold data to provide recommendations for new users, and try to allow users to input less or even no input during the cold start period, thereby reducing the operational complexity of mobile users. In the user data analysis stage, a wealth of contextual information is combined, including environment, weather conditions, time seasons, and the most effective recommendations are provided to users based on their short-term preferences and fixed preferences. When the system accumulates enough data, a collaborative filtering algorithm is applied to improve the recommendation results. The present invention combines situational information and considers the problem of system cold start, and applies a collaborative filtering algorithm and a rule-based recommendation algorithm to greatly improve the accuracy of recommending personalized restaurants for users.

Description

A personalized restaurant recommendation method combined with contextual information

technical field

The invention relates to the technical field of information recommendation, in particular to a personalized restaurant recommendation method combined with context information.

Background technique

With the popularity of the Global Positioning System (GPS) and the rapid development of mobile terminal technology, location-based services (LBS) are more and more widely used in work and daily life environments. The location-based service obtains the location information of the mobile terminal user through the radio communication network (such as GSM network, CDMA network) or external positioning method (such as GPS) of the telecom mobile operator. With the support of the geographic information system platform, it can Users provide appropriate information services. In China, mobile apps such as Dianping, which provide consumers with location-based service recommendations including catering, hotels, tourism, leisure and entertainment, are becoming more and more popular.

At present, most restaurant recommendation systems based on location-based services only focus on analyzing users' fixed preferences to make recommendations, such as the price range, taste, environment of the place, and service level of the restaurant that the user is accustomed to choosing. But with different contextual information, users will have some short-term preferences that are different from fixed preferences. For example, in winter, users are more likely to choose hot pot restaurants instead of outdoor barbecue restaurants; in the morning, users are more likely to choose porridge noodle restaurants. The contextual information that can affect the user's short-term preferences includes the time, location, weather, season, etc. of the user. However, existing recommendation systems do not consider the impact of such contextual information on users' short-term preferences.

Contents of the invention

The purpose of the present invention is to overcome the deficiencies of the prior art, combine situational information, and consider the problem of system cold start, apply collaborative filtering algorithm and rule-based recommendation algorithm, and provide a personalized restaurant with more accurate recommendations combined with situational information recommended method.

In order to achieve the above object, the technical scheme provided by the present invention is:

It is divided into three stages, which are the establishment of the rule base, the cold start stage and the user data analysis stage;

Among them, the established rule base contains short-term preference rules and fixed preference rules;

In the cold start phase, use cold data to provide recommendations for new users, and try to allow users to input less or even no input during the cold start period, thereby reducing the operational complexity of mobile users. The specific steps are: first combine the context information with the rule base Match the short-term preference rules to obtain the short-term preference of the user; then match the user profile with the fixed preference rule in the rule base to obtain the fixed preference of the user; finally, match the obtained short-term preference and fixed preference with the attributes of the restaurant to calculate Probability of restaurant recommendation.

After the cold start phase, a large amount of accumulated user feedback data and interactive data can be used for data mining and analysis of user behavior, that is, to enter the user data analysis phase. The specific steps of this phase are: first modify the rules of the cold start phase, and then conduct based User and scenario-based collaborative filtering algorithm to obtain the collaborative filtering results, and finally the results obtained by the rule-based recommendation algorithm (that is, the recommendation results obtained in the cold start phase) are mixed with the collaborative filtering results to obtain the final recommendation results and Recommend the resulting restaurant to the target user.

Further, the short-term preference rules are initially established based on common sense and are corrected in real time when the user interacts with the personalized restaurant recommendation system; the fixed preference rules are determined by analyzing data. Both rules are (if...then...) statements that are primarily created using static attributes or dynamic information about the user.

Further, in the user data analysis phase, the rules in the cold start phase are modified, specifically: by analyzing historical data, searching for association rules and recalculating the probability of each rule in the rule base.

Further, user-based and scenario-based collaborative filtering algorithms are carried out in the user data analysis stage, and the specific steps for obtaining collaborative filtering results are as follows:

First, determine the neighbor user most similar to the target user and obtain the restaurant selection of similar neighbors to obtain the recommendation result; then add context information to obtain the restaurant selection of similar users in the same situation as the target user to obtain the recommendation result; finally Combine the two recommendation results to get the collaborative filtering result.

In this scenario, the input data includes:

Contextual information, including the user's location, weather, time, and environment;

User personal information, including mobile phone registration information: such as gender, age and other demographic information; and mobile device information: such as operating system, mobile phone model, mobile phone commonly used software, etc.;

Restaurant information, including restaurant category, attributes and some basic information. Restaurant categories include hot pot, Cantonese cuisine, Sichuan cuisine, barbecue, etc. Catering attributes include price, environment, rating, taste, atmosphere, whether to provide wireless network, etc. Basic restaurant information includes location, business hours, phone number, etc.

User logs, which record user ratings and interaction data for restaurants, are used to update the rule support in the cold start phase and provide a data basis for the subsequent collaborative filtering phase.

Compared with the existing technology, the principle and advantages of this scheme are as follows:

This solution combines situational information and considers the problem of system cold start. Using collaborative filtering algorithm and rule-based recommendation algorithm, a three-stage recommendation method is designed, which greatly improves the accuracy of recommending personalized restaurants for users.

Description of drawings

Fig. 1 is a block flow diagram of a personalized restaurant recommendation method combined with contextual information according to the present invention;

Fig. 2 is a flow chart of the cold start stage in a personalized restaurant recommendation method combined with contextual information according to the present invention;

Fig. 3 is a flowchart of collaborative filtering recommendation in a personalized restaurant recommendation method combined with context information according to the present invention.

Detailed ways

The present invention will be further described below in conjunction with specific embodiment:

Referring to Fig. 1 , a personalized restaurant recommendation method combined with contextual information described in this embodiment includes establishing a rule base, a cold start phase, and a user data analysis phase.

Among them, the established rule base contains short-term preference rules and fixed preference rules;

In the cold start phase, due to the lack of historical data and user feedback scores in the cold start phase, coupled with the large number of discrete restaurant attributes, and the low tolerance of mobile end users to complex operations, this embodiment chooses a rule-based algorithm to provide New users provide recommendations, as shown in Figure 2, and the specific process is as follows:

First match the context information with the short-term preference rules in the rule base to obtain the user's short-term preference, specifically:

The user's preference probability for the restaurant of type i in different situations is:

Among them, T _i , W _i and E _i respectively represent the preference probability of restaurants of type i under the current time, weather and environment; i=1, 2, 3..., s, s is the number of restaurant categories;

Assuming that K _i is the preference probability of the restaurant category, the user's short-term preference is expressed as: K ₁ , K ₂ , K ₃ ,...,K _i ; where K _i =1/3(T _i +W _i +E _i ).

After the short-term preference is obtained, the user's fixed preference is obtained; in the process of calculating the user's fixed preference, the fixed preference rules in the rule base are matched with the user's attributes to calculate the preference probability of each restaurant's attribute value. Then the user fixed preference can be expressed as:

Among them, i=1, 2, 3..., m, represents the number of values of each attribute of the restaurant; j=1, 2, 3..., n, represents the number of restaurant attributes; each column in the formula represents the user's opinion of the restaurant The probability of preference for an attribute value for a particular attribute. For example, Table 1 below describes the user's preference probability for each attribute value of the dining atmosphere attribute of a certain restaurant.

Table 1

After obtaining the user's fixed preferences and short-term preferences, the recommendation probability of the restaurant is calculated by matching the above rules with the attributes of the restaurant. For example, the attributes of a certain restaurant and the user preference probability for each specific attribute are shown in Table 2 below:

Table 2

Finally, in the cold start phase, the recommendation probability of the restaurant is calculated by matching the obtained short-term preferences and fixed preferences with the attributes of the restaurant. The calculation formula is as follows:

After passing the cold start phase, enter the user data analysis phase. This phase includes the following steps:

S1. Modify the rules of the cold start phase:

That is, by analyzing historical data, finding association rules and recalculating the probability of each rule in the rule base.

S2. Perform user-based and scenario-based collaborative filtering algorithms to obtain collaborative filtering results, as shown in Figure 3, the specific process is as follows:

First determine the neighbor users who are most similar to the target user and obtain the restaurant choices of similar neighbors to obtain the recommendation results; then add context information to obtain the restaurant choices of similar users in the same situation as the target user to obtain the recommendation results; finally Combine the two recommendation results to get the collaborative filtering result.

S3. Finally, the result obtained by the rule-based recommendation algorithm is mixed with the collaborative filtering result to obtain the final recommendation result and recommend the restaurant of the result to the target user.

In this embodiment, a three-stage recommendation method is designed by combining the situational information and considering the system cold start problem, applying the collaborative filtering algorithm and the rule-based recommendation algorithm, so that the accuracy of recommending personalized restaurants for users is greatly improved.

The implementation examples described above are only preferred embodiments of the present invention, and are not intended to limit the scope of the present invention. Therefore, all changes made according to the shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims (6)

1. method is recommended in a kind of personalized dining room of combination contextual information, it is characterised in that：Including establishing rule base, cold start-up rank Section and Users'Data Analysis stage；

Wherein, short-term preference rules and fixed preference rules are contained in the rule base of foundation；

In cold-start phase, first contextual information is matched with the short-term preference rules in rule base, obtains the short-term preference of user； Subscriber data is matched with the fixation preference rules in rule base again, obtains the fixation preference of user；Finally by what is got Short-term preference and the attribute in fixed preference and dining room, which match, calculates the recommendation probability in dining room；

In the Users'Data Analysis stage, the rule of cold-start phase is changed first, is then carried out based on user and based on scene Collaborative filtering, draws collaborative filtering as a result, the result and collaborative filtering that are finally obtained rule-based proposed algorithm As a result mix, draw consequently recommended result and targeted customer is recommended into the dining room of the result.

2. method is recommended in a kind of personalized dining room of combination contextual information according to claim 1, it is characterised in that：It is described Short-term preference rules, it is initially based on general knowledge foundation, and is carried out when user interacts with personalization dining room commending system Corrigendum in real time；The fixed preference rules are determined by analyzing data.

3. method is recommended in a kind of personalized dining room of combination contextual information according to claim 1, it is characterised in that：It is described User fixes the acquisition of preference in cold-start phase, specific as follows：

Fixation preference rules in the attribute and rule base of subscriber data are matched, calculate the preference of each dining room property value Probability, then show that user fixes preference and is：

Wherein, i=1,2,3 ..., m, represent the quantity of the value of each attribute in dining room；J=1,2,3 ..., n, represent that dining room belongs to The quantity of property；Each row represent preference probability of the user to the property value of dining room particular community in formula.

4. method is recommended in a kind of personalized dining room of combination contextual information according to claim 1, it is characterised in that：It is described Matched in cold-start phase by the short-term preference and the attribute in fixed preference and dining room that get and calculate the recommendation in dining room Probability, calculation formula are as follows：

<mrow> <mi>P</mi> <mo>=</mo> <msub> <mi>K</mi> <mi>i</mi> </msub> <mo>&amp;times;</mo> <mfrac> <mn>1</mn> <mi>n</mi> </mfrac> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>n</mi> </munderover> <msub> <mi>p</mi> <mi>i</mi> </msub> </mrow>

Wherein, K_iFor the preference probability of dining room species, p_iFor the corresponding preference probability of dining room attribute.

5. method is recommended in a kind of personalized dining room of combination contextual information according to claim 1, it is characterised in that：It is described The rule of cold-start phase is changed in the Users'Data Analysis stage, is specially：By analysis of history data, correlation rule is searched simultaneously Recalculate the regular probability of each in rule base.

6. method is recommended in a kind of personalized dining room of combination contextual information according to claim 1, it is characterised in that：It is described The collaborative filtering based on user and based on scene is carried out in the Users'Data Analysis stage, draws the specific of collaborative filtering result Step is：

Determine first with the most like neighbor user of targeted customer and obtain the dining room selection of similar neighborhood, so as to obtain recommending knot Fruit；Then contextual information is added, obtains in selection of the similar users to dining room being in targeted customer under identical situation, obtains Recommendation results；Finally two recommendation results are combined together, draw collaborative filtering result.