选择性接触机制引发的网络极化现象日益严重，同时社交媒体中情感传播的非理性特征进一步加重舆论场的失序。交叉接触作为审议协商的要件，有潜力平衡一致接触带来的态度极化倾向，且社交媒体制造的更多空间间性增加了用户偶然的交叉接触频率。但近来研究发现，交叉接触不仅对态度改变的影响微乎其微，甚至会出现回旋镖效应反而进一步加重极化现象。有关交叉接触与态度极化的复杂关系及其影响机制和影响因素，尚未得到充分探究。加之，大部分极化研究都集中在政治、党派或意识形态方面，但国内网络极化现象较少受到党派因素影响。因而，本文将立足我国政治语境，探究多元社会文化环境中极化现象。

在此背景下，本研究基于民主协商理论、情智理论和认知加工偏见机制，从微观视角出发，考察交叉接触如何通过说服机制影响态度极化，并在此基础上进一步探究其背后深一层次的影响机制问题，即广泛存在于舆论场的愤怒和恐惧两种离散情绪是否会对其产生不同的调节作用，以及论据加工偏向是否在其中发挥着中介作用。

本研究通过两项研究分别对两个层次的研究问题及研究假设进行了考察和检验。研究一（N = 154）发现，交叉接触组参与者的态度极化程度显著低于一致接触组。研究二（N = 290）发现，相比控制组，愤怒情感下交叉接触的态度缓和作用相对较弱，恐惧情感下这一作用相对较强；论据加工偏向的中介作用并不存在，尽管论据加工偏向的确会促使态度极化，但交叉接触与论据加工偏向无显著相关。

本研究结果表明，首先，社交媒体中交叉接触推动协商讨论的积极作用仍然存在。其次，情感对其产生影响。具体而言，愤怒的封闭式认知特性不利于态度缓和，而恐惧的开放式认知特性则有利于培养协商意愿较强且态度温和的公民。这一研究结果揭示了两种离散情感对于网络协商民主过程的不同影响，并进一步厘清公共领域中“好的情感”与“坏的情感”。最后，交叉接触对论据评价影响的复杂性表明，在对交叉信息的论据质量感知中，协商理论中的平等评价与常见的失证偏差可能同时存在。

总体上，本研究对认识非党派环境下社交媒体中极化现象具有一定的启示意义。不过，本研究仅选择了一个议题作为实验材料，议题类型较为有限，未来研究可在更多元化的议题中进行考察，从而验证本研究结果的普遍性。

Network polarization caused by the selective exposure mechanism is becoming increasingly serious, and the irrational character of emotional communication on social media further aggravates the disorder of the public opinion field. Cross-cutting exposure, as an element of deliberation, has the potential to balance the attitude polarization tendency brought about by like-minded exposure, while social media increases the frequency of users' accidental cross-cutting exposure by creating more interspatiality. But recent research has found that cross-cutting exposure has little effect on attitude change and even a boomerang effect that further exacerbates polarization. This complex relationship between cross-cutting exposure and attitude polarization has not been fully explored, as well as its mechanism and moderating factors. In addition, most of the studies concerning polarization focus on politics, partisanship, or ideology, whereas domestic network polarization is barely affected by those factors. Therefore, the present study will be based on the political context of China and explore polarization in a diverse social and cultural environment.

In this context, based on democratic deliberation theory, affective intelligence theory, and cognitive processing bias mechanism, this study examines how cross-cutting exposure affects attitude polarization through persuasion mechanism from a micro perspective. On this basis, the present study further explores the problem of the next level, that is, whether the two discrete emotions of anger and fear, that are widespread in public opinion, have different moderating effects on it and whether the argument processing bias plays a mediating role in it.

This study examines and tests two levels of research questions and research hypotheses through two studies. The experimental results of Study 1 (N = 154) found that the attitude polarization of participants in the cross-cutting exposure group was significantly lower than that in the like-minded exposure group. Study 2 (N = 290) found that compared with the control group, the attitude moderation effect of cross-cutting exposure in the anger group was relatively weak, and this effect was relatively strong in the fear group; argument processing bias was not a significant mediating role, although argument processing bias did promote attitude polarization, and cross-cutting exposure was not significantly associated with argument processing bias.

The results of this study suggest that, first of all, cross-cutting exposure remains to play a positive role in deliberations in social media. Secondly, emotion affects this relationship. Specifically, the closed cognitive pattern of anger is not conducive to moderation while the open cognitive pattern of fear supports cultivates citizens with a strong willingness to deliberate and moderate attitudes. This research reveals the distinct effect of two discrete emotions on the process of online deliberative democracy and further clarifies the "good emotions" and "bad emotions" in the public sphere. Thirdly, the complexity of the impact of cross-cutting exposure on argument evaluation indicates that equal evaluation in deliberation theory might coexist with the common disconfirmation bias in the perception of argument quality for cross-cutting information.

Overall, this study has certain implications for enriching knowledge of social media polarization in a non-partisan setting. However, the range of topics is relatively constrained, as there was only one topic selected as the experimental material. Future research can investigate a wider range of issues, so as to verify if the findings of this study are applicable to general situation.